WO2020228739A1 - 光伏组件缺陷的检测方法和装置、分类器的训练方法、终端设备及非暂时性存储介质 - Google Patents

光伏组件缺陷的检测方法和装置、分类器的训练方法、终端设备及非暂时性存储介质 Download PDF

Info

Publication number
WO2020228739A1
WO2020228739A1 PCT/CN2020/090032 CN2020090032W WO2020228739A1 WO 2020228739 A1 WO2020228739 A1 WO 2020228739A1 CN 2020090032 W CN2020090032 W CN 2020090032W WO 2020228739 A1 WO2020228739 A1 WO 2020228739A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
area
information
training
classifier
Prior art date
Application number
PCT/CN2020/090032
Other languages
English (en)
French (fr)
Inventor
张欢欢
唐小军
李慧
Original Assignee
京东方科技集团股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 京东方科技集团股份有限公司 filed Critical 京东方科技集团股份有限公司
Publication of WO2020228739A1 publication Critical patent/WO2020228739A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02S—GENERATION OF ELECTRIC POWER BY CONVERSION OF INFRARED RADIATION, VISIBLE LIGHT OR ULTRAVIOLET LIGHT, e.g. USING PHOTOVOLTAIC [PV] MODULES
    • H02S50/00—Monitoring or testing of PV systems, e.g. load balancing or fault identification
    • H02S50/10—Testing of PV devices, e.g. of PV modules or single PV cells
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G06T2207/30164—Workpiece; Machine component
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E10/00—Energy generation through renewable energy sources
    • Y02E10/50—Photovoltaic [PV] energy

Definitions

  • the embodiments of the present disclosure relate to a detection method and device for photovoltaic module defects, a training method for a classifier, terminal equipment, and a non-transitory storage medium.
  • photovoltaic power generation has become the most widely used form of new energy power generation among various new energy sources.
  • the core component of the entire photovoltaic power station is the solar photovoltaic panel. Defects inevitable in the production and installation process will affect its work efficiency. Therefore, it is very necessary to detect the defects of the solar photovoltaic panel.
  • At least one embodiment of the present disclosure provides a method for detecting defects of photovoltaic modules, including: acquiring a target image of a photovoltaic module to be inspected; detecting a salient area of the target image to obtain position information of at least one salient area; The location information of one salient area obtains at least one input image; the at least one input image is input to the classifier, so that the classifier outputs the category information of the at least one salient area.
  • the category information is category information of the defect included in the at least one salient area or identification information of a normal area.
  • the target image is a visible light image.
  • the classifier is trained, and the training method of the classifier includes: inputting a partial image extracted from a sample image as a training sample into the classifier to be trained so that the classifier to be trained Outputting training category information of the partial image, the partial image including a partial region in the sample image, and the training category information being defect category information or normal region identification information included in the partial image; and Determine whether to adjust the classifier to be trained based on the training category information.
  • the training method of the classifier further includes: extracting the partial image from the corresponding position of the sample image according to the position information of the partial image.
  • extracting the partial image from the corresponding position of the sample image according to the position information of the partial image includes: labeling the sample image to obtain a labeling frame including the local area and generating the labeling frame Location information.
  • extracting the partial image from the corresponding position of the sample image according to the position information of the partial image further includes: centering the center point of the labeling frame and taking the long sides of the labeling frame as sides. Extract the square partial area from the corresponding position in the sample image as the partial image.
  • the color information of each pixel in the target image includes RGB color information
  • the detecting the salient area of the target image to obtain the position information of the at least one salient area includes: Gaussian smoothing is performed on the target image to obtain a new target image; the RGB color information of each pixel in the new target image is mapped from the current RGB color space to the CIELab color space to obtain CIELab color information; and the new target image is calculated
  • the respective channel averages of the L, a, and b color channels of all pixels in the pixel for each color channel, calculate the Euclidean distance between the color channel value of each pixel in the new target image and the channel average to obtain the significance Figure; normalize the saliency map to obtain a normalized saliency map; perform threshold segmentation on the normalized saliency map to obtain a saliency binary image; according to the saliency binary image
  • the location of the at least one connected area determines the location information of the at least one salient area.
  • the detection method further includes: if the category information is defect category information, using the location information and category information of the salient area to improve the manufacturing process of the photovoltaic module.
  • the defect included in the at least one salient area is an appearance defect of the photovoltaic module.
  • the appearance defects of the photovoltaic module include one of the following: stains, scratches, delamination, chipping, chipped corners, ribbon offset, debris, perforations, and glass bubbles.
  • At least one embodiment of the present disclosure further provides a method for training a classifier, including: inputting a partial image extracted from a sample image as a training sample into the classifier to output the partial image by the classifier.
  • Category information wherein the partial image includes a partial region in the sample image; determining whether to adjust the classifier based on the training category information.
  • the partial image is an image including a single defect or an image including a normal area
  • the training category information is category information of a single defect included in the partial image or identification information of the normal area.
  • the training method further includes: extracting the partial image from a corresponding position of the sample image according to the position information of the partial image.
  • the training method further includes: processing the sample image to obtain position information of the partial image.
  • processing the sample image to obtain the information of the partial image includes: annotating the sample image to obtain a labeling frame including the local area and generating position information of the labeling frame as the The location information of the partial image.
  • extracting the partial image from the corresponding position of the sample image according to the position information of the partial image includes: taking the center point of the labeling frame as the center and taking the long side of the labeling frame as the side length The partial area of the square is extracted from the corresponding position in the sample image as the partial image.
  • At least some embodiments of the present disclosure also provide a photovoltaic module defect detection device, including: a first acquisition circuit for acquiring a target image of the photovoltaic module to be tested; a second acquisition circuit for performing a significant area of the target image Detection to obtain the position information of the salient area; a third acquisition circuit for obtaining an input image according to the position information of the salient area; a classification circuit for inputting the input image into the classifier to be used by the classifier Output category information of the salient area in the target image; the category information is defect category information or normal area identification information.
  • At least some embodiments of the present disclosure also provide a terminal device including a processor and a memory.
  • Computer program instructions are stored in the memory, and the computer program instructions, when run by the processor, execute the detection method or training method as described above.
  • At least some embodiments of the present disclosure also provide a non-transitory storage medium that stores computer program instructions non-temporarily, and when the computer program instructions are executed by a computer, the foregoing detection method or training method can be implemented.
  • Fig. 1 is one of the flowcharts of a method for detecting defects of photovoltaic modules according to an embodiment of the present disclosure
  • FIGS. 2-8 are effect diagrams showing the appearance defects of photovoltaic modules according to embodiments of the present disclosure.
  • FIG. 9 is the second flowchart of the photovoltaic module defect detection method according to an embodiment of the present disclosure.
  • FIG. 10 is the third flowchart of the photovoltaic module defect detection method according to an embodiment of the present disclosure.
  • Fig. 11 is a fourth flowchart of the method for detecting defects of photovoltaic modules according to an embodiment of the present disclosure
  • Fig. 12 is a block diagram of a photovoltaic module defect detection device according to an embodiment of the present disclosure.
  • Fig. 13 is a schematic structural diagram of a terminal device according to an embodiment of the present disclosure.
  • FIG. 14 is a schematic diagram of a storage medium provided by at least one embodiment of the present disclosure.
  • FIG. 15 is a schematic diagram of a photovoltaic module defect detection system provided by at least one embodiment of the present disclosure.
  • FIG. 16 is a schematic structural diagram of a photovoltaic module defect detection device provided by at least one embodiment of the present disclosure.
  • Fig. 1 shows a method for detecting defects of photovoltaic modules according to an embodiment of the present disclosure.
  • the method for detecting defects of photovoltaic modules can be applied to terminal equipment, which can be a personal computer or a server.
  • the method for detecting defects of photovoltaic modules, as shown in Figure 1, may include the following steps 101-104:
  • a target image of the photovoltaic component to be tested is acquired.
  • the target image of the photovoltaic component to be tested can be acquired from the image acquisition device, or the target image of the photovoltaic component to be tested previously stored can be acquired from other equipment (such as PC, server) or storage device (such as hard disk, U disk, etc.) .
  • the image acquisition device may include an industrial camera that takes a picture of the photovoltaic component to acquire the target image.
  • the visible light image of the photovoltaic component to be inspected can be acquired as the aforementioned target image.
  • the categories of appearance defects of photovoltaic modules to be inspected may include:
  • any thin film layer in the effective working area of the photovoltaic module has a gap of more than 10% of the area of a cell, which is visible to the human eye; for example, the effective working area of the photovoltaic module is capable of receiving light and being able to receive light. The area where light is converted into electrical signals;
  • a continuous bubble or peeling layer is formed between the edge of the photovoltaic module and any part of the circuit
  • FIG. 2-8 typical appearance defects of photovoltaic modules can be shown in Figure 2-8.
  • the appearance defect shown in Fig. 2 is the stain/scratch of the battery
  • the appearance defect shown in Fig. 3 is de-crystallizing
  • the appearance defect shown in Fig. 4 is chipping/cutting
  • the appearance defect shown in Fig. 5 is The ribbon is offset
  • the appearance defect shown in Fig. 6 is a chip/perforation
  • the appearance defect shown in Fig. 7 is glass bubbles
  • the appearance defect shown in Fig. 8 is glass scratches.
  • the appearance defects of the photovoltaic module include at least one of the following: stains, scratches, delamination, chipping, chipped corners, ribbon offset, debris, perforations, and glass bubbles.
  • a salient area detection is performed on the target image to obtain position information of at least one salient area.
  • the position information of the salient area is the position information of the salient area in the target image.
  • the position information includes the coordinates of a vertex of the salient area (for example, the upper left corner vertex), and the length (for example, the maximum length) and width (for example, the maximum width) information of the salient area, so that the position information can be used to determine the The location of the salient area.
  • the salient area is a rectangular area with the point determined by the coordinates as the corresponding vertex, and the maximum length and maximum width as the length and width, respectively.
  • the saliency detection is the visual saliency detection (Visual Saliency Detection), which refers to the use of intelligent algorithms to simulate human visual characteristics to extract the salient regions (regions of interest to humans) in the target image.
  • Visual Saliency Detection refers to the use of intelligent algorithms to simulate human visual characteristics to extract the salient regions (regions of interest to humans) in the target image.
  • humans When facing a scene (for example, the target image), humans automatically process regions of interest and selectively ignore regions that are not of interest. These regions of interest to people are called salient regions.
  • the color, brightness, edge and other characteristics of the salient area are different from the surrounding pixels.
  • the salient area in the target image may be an area in the target image that is different in content from most areas.
  • the salient area in the target image may also be a normal area without appearance defects.
  • the defect detection method of the photovoltaic module obtaineds the image of the suspected defect by first detecting the significant area of the target image, which has at least the following three advantages:
  • each defect area of the target image is formed into a single image and input into the classifier. Compared with directly inputting the entire target image into the classifier or detector, the calculation speed and accuracy are further improved;
  • the detection method provided by the embodiments of the present disclosure can help achieve different The balance of the number of samples of the defect category helps to improve the comprehensive detection ability of the classifier, thereby improving the accuracy of detection.
  • step 102 may include the following steps 901-907:
  • step 901 Gaussian smoothing is performed on the target image to obtain a new target image.
  • a Gaussian function can be used to perform Gaussian smoothing processing on the target image to obtain a new target image.
  • Gaussian smoothing can also be called Gaussian filtering.
  • Gaussian filtering is the weighted average process of the entire target image. The value of each pixel in the new target image is obtained by weighted average of itself and other pixel values in the neighborhood.
  • a Gaussian convolution kernel with a size of 5*5 can be used to perform Gaussian smoothing on the target image, and the weighted average gray value of pixels in the neighborhood determined by the Gaussian convolution kernel can be used to replace the center of the Gaussian convolution kernel. The value of the pixel point to get the new target image. It should be noted that the size of the Gaussian convolution kernel may not be limited to 5*5.
  • the color information of each pixel in the target image may include RGB color information.
  • the color information of each pixel in the target image may include a red sub-pixel value R, a green sub-pixel value G, and a blue sub-pixel value B.
  • the color information of each pixel in the target image may also include YUV color information, and the YUV color space and the RGB color space can be mutually converted. For example, you can first convert the YUV color space to RGB color space, and then implement the following steps.
  • step 902 the RGB color information of each pixel in the new target image is mapped from the current RGB color space to the CIELab color space to obtain the CIELab color information.
  • the RGB color information of each pixel in the new target image can be mapped from the current RGB color space to the XYZ color space to obtain the XYZ information.
  • the following formula (1) can be used to calculate Get XYZ information:
  • M is a preset matrix, and the value of M can be as shown in the following formula (2):
  • the XYZ information of each pixel in the new target image is converted to the CIELab color space to obtain the CIELab color information, which can be calculated using the following formulas (3)-(6):
  • Xn, Yn, and Zn can all be 1.
  • the value range of X, Y, Z and t variables can all be [0,1], the value range of the corresponding L component is [0,100], the value range of a and b components can both be [-127 , 127].
  • step 903 the respective channel average values of the L, a, and b color channels of all pixels in the new target image are calculated.
  • the channel average value of the L color channel can be calculated according to the color channel values of the L color channel of all pixels in the new target image
  • the channel average value of the a color channel can be calculated according to the color channel value of the a color channel of all pixels in the new target image
  • the average value of the b color channel can be calculated according to the color channel value of the b color channel of all pixels in the new target image
  • step 904 for each color channel, the Euclidean distance between the color channel value of each pixel in the new target image and the channel average value is calculated to obtain a saliency map.
  • the L color channel calculates the L color channel value and the channel average value of each pixel in the new target image
  • the a color channel calculate the a color channel value and the channel average value of each pixel in the new target image
  • the b color channel calculate the b color channel value and the average value of each pixel in the new target image Euclidean distance to get a saliency map.
  • step 905 the saliency map is normalized to obtain a normalized saliency map.
  • step 906 threshold segmentation is performed on the normalized saliency map to obtain a saliency binary image.
  • the maximum between-class variance method can be used to perform adaptive threshold segmentation on the saliency map, and the saliency binary image. It should be noted that in practical applications, other methods can also be used to threshold the normalized saliency map to obtain a saliency binary image, which is not limited to the maximum between-class variance method (OTSU) provided in the embodiments of the present disclosure. ).
  • the location information of the at least one salient area is determined according to the location of the at least one connected area in the salient binary image.
  • the position information includes the coordinates of a vertex of the salient area (for example, the upper left corner vertex), and the maximum length and the maximum width involved in the salient area, so that the location of the salient area can be determined according to the position information.
  • the salient area is a rectangular area with the point determined by the coordinates as the corresponding vertex, and the maximum length and maximum width as the length and width, respectively.
  • the saliency binary image may have one, two, or more than two connected regions.
  • the position of each connected area is the position of the salient area.
  • the binary image includes a first pixel value area and a second pixel value area.
  • the pixel value of the first pixel value area is 1, which is a black area and corresponds to a normal area; the pixel value of the second pixel value area If it is 0, it is a white area, corresponding to a salient area, and the area covered by multiple white pixels adjacent to each other is recognized as a salient area. Therefore, the position information of at least one salient area can be determined according to the position of at least one connected area in the salient binary image.
  • the location information of at least one connected area may be correspondingly determined as the location information of at least one salient area, or the location information of at least one connected area may be fine-tuned, and then the location information of the at least one connected area after the fine-tuning may be determined Is the location information of at least one salient area, but not limited to this.
  • the edge of a part or each connected area can be expanded outward by a specified number of pixels to obtain an enlarged connected area, and then the location information of the enlarged connected area can be determined Is the location information of the salient area.
  • the specified number can be 1-3, but is not limited to this. In this way, the salient area can include more information, which helps improve the accuracy of defect recognition.
  • the target image may also be input to the salient area detector, so that the salient area detector outputs the position information of the salient area.
  • the salient area detector recognizes the salient area and obtains position information of the salient area.
  • the salient area detector can be implemented in an appropriate manner, such as a saliency detection neural network; for example, deep learning technology can be used to train the salient area detector.
  • an algorithm can refer to Zhao T, Wu X. Pyramid Feature Selective Network for Saliency detection[C]. 2019 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2019.
  • the algorithm uses a context-aware pyramid attention network
  • a saliency map is generated, where the pyramid network is based on VGG-16's conv3-3, conv4-3 and conv5-3 as the basic network, using convolutions with different expansion rates (such as 3, 5, and 7) to capture contextual information of multiple perception fields.
  • convolutions with different expansion rates such as 3, 5, and 7.
  • CA channel attention network
  • SA spatial attention network
  • step 103 at least one input image is obtained according to the location information of the at least one salient area.
  • At least one salient area can be extracted from the target image according to the location information of the at least one salient area to obtain at least one input image, and the input image is the image of the salient area.
  • the salient area extracted from the target image can be used as the input image.
  • At least one salient area can be extracted from the target image by the following method to obtain at least one input image: for the location information corresponding to each salient area, the corresponding salient area is determined, and for each salient area, The long side of the circumscribed rectangle of the salient area is taken as the side length, and the square area is extracted with the center of the circumscribed rectangle as the center to obtain the input image.
  • the shape or size of the input image can meet the requirements of the classifier for the input image.
  • step 104 the at least one input image is input to a classifier, so that the classifier outputs category information of at least one salient area in the target image; the category information is defect category information or normal area
  • the identification information that is, background information. Among them, there are no defects in the normal area.
  • the classifier may be pre-trained, may be pre-stored in the terminal device or stored in an external device of the terminal device, and be called when needed.
  • the classifier After inputting at least one input image into the trained classifier, for each input image, the classifier identifies the category information of the corresponding salient area. After the classifier outputs the class information of the salient area, if the salient area The category information of is the category information of the defect, and the location information of the defect area in the target image can be determined according to the location information of the salient area. Of course, if the classifier determines that the salient area is a normal area, it will also output the identification information of the normal area. After obtaining the category information of the salient area corresponding to each input image, the terminal device can store the recognition result and determine whether all input images have been recognized. If not, control the classifier to perform defect recognition on the next input image. , The location information and defect category information of all defect areas in the target image are output.
  • the trained classifier can identify two defect categories, a binary classification algorithm can be used to identify the defect category, and if the trained classifier can identify more than two defect categories, then multi-classification can be used. Algorithms to identify the category of the defect, for example, it can identify whether the defect category in the input image is cell stains, chipping, or glass bubbles.
  • the training method of the above-mentioned classifier may at least include the following steps 1001-1002 training:
  • a partial image extracted from a sample image is input as a training sample into a classifier to be trained so that the classifier to be trained outputs training category information of the partial image, wherein the partial image includes the For a partial area in the sample image, the training category information is defect category information or normal area identification information included in the partial image.
  • step 1002 it is determined whether to adjust the classifier to be trained based on the training category information.
  • the training samples can be input to the classifier to be trained, and the classifier is trained to obtain the trained classifier.
  • Each training sample has a preset training category information corresponding to it, and the training category information of the training sample may include defect category information or negative sample identification information.
  • the corresponding training category information can be preset to 1; when the local area included in the training sample has chipped edges, the corresponding training category information can be preset to 2; training When there are glass bubbles in the local area included in the sample, the corresponding training category information can be preset to 3; when the local area included in the training sample is a normal area (area without defects), that is, the training sample is a negative sample At this time, the corresponding training category information can be preset to 4.
  • the output training category information does not match the preset corresponding training category information of the training sample, it is determined to adjust the classifier (such as adjusting the parameters of the classifier) until the output training category information corresponds to the preset Match the training category information, and enter the training of the next training sample.
  • the classifier such as adjusting the parameters of the classifier
  • the partial image is an image that includes a defect and includes a single defect, that is, only includes one type of defect, which can increase the feature saliency of the defect, thereby improving the judgment ability of the classifier. Since in the detection method provided by the embodiment of the present disclosure, a partial image is extracted from the target image as a training sample, the region with a single defect can be extracted in a targeted manner, and the selection of training samples is more flexible.
  • a deep learning classification technique can be used to train the classifier.
  • ResNet residual Neural Network
  • the main idea of ResNet is to add a direct connection channel to the network, that is, the idea of Highway Network (high-speed road neural network).
  • Highway Network high-speed road neural network
  • the previous network structure is a nonlinear transformation of performance input, while Highway Network allows to retain a certain proportion of the output of the previous network layer.
  • the idea of ResNet is very similar to that of Highway Network, allowing the original input information to be directly transmitted to the subsequent layers.
  • two residual modules are used in the ResNet network structure.
  • One is a residual module with two 3*3 convolutional networks connected in series, and the other is 1*1, 3 *3.
  • 3 convolutional networks of 1*1 are connected in series as a residual module.
  • ResNet can have different network layers.
  • the number of network layers in ResNet can be 50-layer (layer), 101-layer or 152-layer.
  • ResNets with different network layers are all implemented by stacking the above residual modules.
  • the training method of the classifier may further include: extracting the partial image from the corresponding position of the sample image as the training sample according to the position information of the partial image.
  • the above-mentioned training samples can be obtained through the following steps 1101-1104:
  • step 1101 the sample image is annotated to obtain an annotation frame including the local area, and position information of the annotation frame is generated as the position information of the partial image.
  • the sample image can be manually annotated, the defective area or the normal area can be framed with the annotation frame, and the position information of the annotation frame can be generated as the position information of the partial image, and the training corresponding to the annotation frame can be added at the same time.
  • the category information is used to identify the defect category in the area framed by the labeling frame, or to identify negative samples.
  • the corresponding training category information may be defect category information
  • the training category information may be identification information of a negative sample.
  • the sample position information includes the coordinates of a vertex of the labeling frame (for example, the upper left corner vertex), and the length and width of the labeling frame.
  • the corresponding training category information when there are battery stains in the area framed by the label box, you can add the corresponding training category information as 1; when the area framed by the label box has collapsed edges, you can add the corresponding training category information as 2; When there are glass bubbles in the area, the corresponding training category information can be added as 3; when the area enclosed by the label box is a normal area, the corresponding training category information can be added as 4.
  • target detection and labeling tool software can be used to label the sample image.
  • the target detection and labeling tool software may be LabelImage software, for example.
  • detection and labeling tool software to label images has the advantages of high accuracy, which can improve the accuracy of the information carried by the sample image, thereby obtaining high-quality training samples, and thereby improving the accuracy of the detection results.
  • multiple normal regions with different positions or different shapes can be selected to form multiple negative sample images to train the classifier, which can improve the accuracy of detection.
  • a square local area is extracted from a corresponding position in the sample image with the center point of the label frame as the center and the long side of the label frame as the side length as the local image.
  • partial images can be used as original training samples.
  • the original training sample can be used as an information source, and data is enhanced through image processing to obtain multiple derived training samples (derived training samples).
  • step 1103 data enhancement is performed on the training samples to obtain multiple derived training samples.
  • image processing such as rotation and scaling can be performed on the original training samples to achieve data enhancement, and a series of training samples can be obtained.
  • the original training samples When the original training samples are rotated, the original training samples can be rotated by a specified angle in turn to obtain a series of derivative training samples.
  • scaling the original training samples the original training samples may be reduced by a specified multiple in turn, and the original training samples may be enlarged by a specified multiple in turn to obtain a series of derivative training samples. In this way, the diversity of training samples can be increased, which is conducive to improving the accuracy of the classifier.
  • the original training sample and the corresponding derived training sample have the same label.
  • the foregoing original training samples and multiple derived training samples obtained through data enhancement can be input into the classifier to be trained, and the classifier is trained to obtain the trained classifier.
  • step 1104 a label of a training sample is generated according to the training category information. This label is used for input into the classifier.
  • the label of the training sample can be generated according to the training category information corresponding to the label box of the training sample. For example, when the training category information corresponding to the label box of the training sample is battery stain, the label of the training sample is 1, and when the training category information corresponding to the label box of the training sample is edge collapse, the label of the training sample is 2. When the training category information corresponding to the label box of the training sample is a glass bubble, the label of the training sample is 3; when the training sample is a negative sample, the label of the training sample is 4.
  • the detection method may further include: if the category information output in step S104 is defect category information, using the location information and category information of the salient area to improve the manufacturing process of the photovoltaic module.
  • the category information output in step S104 is the category information of the defect, it means that there is a defect in the target image, and the location information and category information of the salient area are the location information and category information of the defect.
  • the defect detection method of the photovoltaic module provided by the embodiment of the present disclosure can locate the defect on the photovoltaic module by obtaining the position information of the defect, which not only meets the requirements of quality inspection, but also provides more effective information.
  • the defect is a stain and occurs in an ineffective working area of the photovoltaic module, and the defect will not affect the normal operation of the photovoltaic module, it can be judged that the defect is a tolerable defect.
  • the location information and category information of the defect can be analyzed, such as big data analysis.
  • the analysis results show that the probability of a specific defect occurring in a certain position of the photovoltaic module is high. It can be considered to improve the related process steps of the formation of the defect, so as to provide a targeted process for the process. Feedback information is conducive to the optimization of the process flow.
  • the salient area detection is performed on the target image of the photovoltaic module to be inspected first, at least one salient area is obtained as a candidate area of the defect area, and then at least one input image is obtained according to the position information corresponding to the at least one salient area , And input at least one of the above-mentioned input images into the classifier, so that the classifier outputs the category information of at least one salient area in the target image, so that the user can know whether the photovoltaic module to be inspected has defects, and when there are defects, the defect category .
  • the aforementioned category information is defect category information or normal area identification information. In this way, the amount of image processing data can be reduced, thereby improving the detection efficiency of photovoltaic module defects and realizing real-time detection.
  • the embodiments of the present disclosure turn the problem of target detection into a problem of saliency area detection and target recognition, which not only ensures the accuracy of the algorithm but also ensures the real-time performance of the algorithm.
  • the embodiment of the present disclosure also proposes a photovoltaic module defect detection device. As shown in FIG. 12, the device includes:
  • the first acquisition module 121 is configured to acquire a target image of the photovoltaic component to be tested
  • the second acquisition module 122 is configured to perform salient area detection on the target image to obtain location information corresponding to at least one salient area;
  • the third obtaining module 123 is configured to obtain at least one input image according to the location information corresponding to the at least one salient area;
  • the classification module 124 is configured to input the at least one input image into a classifier, so that the classifier outputs category information of at least one salient area in the target image; the category information is defect category information or normal Identification information of the area.
  • the first acquisition module 121, the second acquisition module 122, the third acquisition module 123, and the fourth classification module 124 can be implemented by hardware (for example, circuit) modules, software modules, or any combination thereof, which is not limited in the present disclosure.
  • the salient area detection is performed on the target image of the photovoltaic module to be inspected first, at least one salient area is obtained as a candidate area of the defect area, and then at least one salient area is obtained according to the position information corresponding to the at least one salient area Input an image, and input at least one of the above-mentioned input images into the classifier, so that the classifier outputs the category information of at least one salient area in the target image, so that the user can know whether the photovoltaic module to be inspected has defects, and when there are defects Category.
  • the aforementioned category information is defect category information or normal area identification information. In this way, the amount of image processing data can be reduced, thereby improving the detection efficiency of photovoltaic module defects and realizing real-time detection.
  • At least one embodiment of the present disclosure further provides a method for training a classifier, including: inputting a partial image extracted from a sample image as a training sample into the classifier to output the partial image by the classifier.
  • Category information wherein the partial image includes a partial region in the sample image; and determining whether to adjust the classifier based on the training category information.
  • a partial image is extracted from the target image as a training sample. For example, compared to directly inputting the entire target image to the classifier or detector, it has the following at least three advantages :
  • the data volume of a single training sample is reduced, thereby reducing the computing power requirements of the classifier and increasing the computing speed;
  • Each training sample has a preset corresponding training category information. For example, if the output training category information does not match the preset corresponding training category information of the training sample, it is determined to adjust the classifier (for example, adjust the classifier Parameter) until the output training category information matches the preset corresponding training category information, and the training of the next training sample is entered.
  • the classifier for example, adjust the classifier Parameter
  • the partial image is an image including a single defect or an image including a normal area
  • the training category information is category information of a single defect included in the partial image or identification information of the normal area.
  • the corresponding training category information can be preset to 1; when the local area included in the training sample has edge collapse, it can be preset Suppose the corresponding training category information is 2; when glass bubbles exist in the local area included in the training sample, the corresponding training category information can be preset to 3; the local area included in the training sample is a normal area (area without defects) When the training sample is a negative sample, the corresponding training category information can be preset to 4.
  • the partial image is an image including a defect or an image including a normal area
  • the training category information is category information of the defect included in the partial image or identification information of the normal area.
  • the partial image is an image that includes a defect and includes a single defect, that is, only includes one type of defect, which can increase the feature saliency of the defect, thereby improving the judgment ability of the classifier. Since in the training method of the classifier provided by the embodiments of the present disclosure, a partial image is extracted from the target image as the training sample, the region with a single defect can be extracted in a targeted manner. In the selection of training samples, compared with The whole target image is more flexible as a training sample.
  • the partial image is extracted from the corresponding position of the sample image according to the position information of the partial image.
  • the sample image is processed to obtain the position information of the partial image.
  • the sample image may be annotated to obtain an annotation frame including the local area, and the position information of the annotation frame may be generated as the position information of the partial image; for example, the position information of the annotation frame includes The coordinates of a vertex (for example, the vertex at the upper left corner) of the label box, and the length and width of the label box.
  • target detection and labeling tool software can be used to label the sample image.
  • the target detection and labeling tool software may be LabelImage software, for example.
  • the classifier used in the photovoltaic module defect detection method provided by at least one embodiment of the present disclosure may be obtained by training the above-mentioned classifier training method.
  • Fig. 13 is a block diagram showing a terminal device according to an exemplary embodiment.
  • the terminal device 1400 may be provided as a server, but it is not limited thereto.
  • the device 1400 includes a processing component 1422, which further includes one or more processors, and a memory resource represented by a memory 1432, for storing instructions executable by the processing component 1422, such as application programs.
  • the application program stored in the memory 1432 may include one or more modules each corresponding to a set of instructions.
  • the processing component 1422 is configured to execute instructions to execute the aforementioned detection method for photovoltaic component defects.
  • the processing component 1422 and the memory 1432 are connected through a bus system.
  • the bus system may be a commonly used serial or parallel communication bus, etc., which is not limited in the embodiments of the present disclosure.
  • the device 1400 may also include a power component 1426 configured to perform power management of the device 1400, a wired or wireless network interface 1450 configured to connect the device 1400 to the network, and an input output (I/O) interface 1458.
  • the device 1400 can operate based on an operating system stored in the memory 1432, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.
  • the processing component 1422 may be a central processing unit (CPU), a digital signal processor (DSP), an image processor (GPU), or other forms of processing units with data processing capabilities and/or instruction execution capabilities, and may be general-purpose A processor or a dedicated processor, and can control other components in the display processing apparatus 200 to perform desired functions.
  • the processor may be a general-purpose processor or a special-purpose processor, and may be a processor based on the X86 or ARM architecture.
  • the memory 1432 may include one or more computer program products, and the computer program products may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory.
  • the volatile memory may include random access memory (RAM) and/or cache memory (cache), for example.
  • the non-volatile memory may include read-only memory (ROM), hard disk, flash memory, etc., for example.
  • One or more computer program instructions can be stored on the computer-readable storage medium, and the processing component 1422 can run the program instructions to implement the functions (implemented by the processing component 1422) and/or other desired functions in the embodiments of the present disclosure, For example, the detection method of photovoltaic module defects.
  • Various application programs and various data such as various data used and/or generated by the application program, can also be stored in the computer-readable storage medium.
  • a non-transitory computer-readable storage medium including instructions is also provided.
  • the storage medium 600 non-transitory stores computer program instructions 601.
  • the computer program instructions 601 are executed by a computer, one or more steps in the photovoltaic module defect detection method described above can be executed, or one or more of the above-described classifier training methods can be executed. Steps.
  • the storage medium 600 may be the foregoing memory 1432 including instructions, and the foregoing instructions may be executed by the processing component 1422 of the device 1400 to complete the foregoing detection method or training method.
  • the non-transitory computer-readable storage medium may be ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.
  • Figure 15 is a system that can be used to implement the photovoltaic module defect detection method provided by the embodiments of the present disclosure.
  • the system 200 may include a user terminal 110, a network 120, a server 130, a database 140, and an image acquisition device 150.
  • the system 200 can be used to implement the photovoltaic module defect detection method provided by any embodiment of the present disclosure, and its specific structure and function can refer to the corresponding content of the system used to implement the photovoltaic module defect detection method below.
  • the user terminal 110 is, for example, a computer 110-1 or a mobile phone 110-2. It is understandable that the user terminal 110 may be any other type of electronic device capable of performing data processing, which may include, but is not limited to, a desktop computer, a notebook computer, a tablet computer, a smart phone, a smart home device, a wearable device, and in-vehicle electronics. Equipment, monitoring equipment, etc. The user terminal 110 may also be any equipment provided with electronic equipment, such as a vehicle, a robot, and the like.
  • the image acquisition device 150 may include a camera.
  • the image acquisition device 150 takes a picture of the photovoltaic component to be detected, and uploads the photographed image data of the photovoltaic component to the server 130 via the network 120 for the user terminal 110 to call. Of course, it can also Upload directly to the user terminal 110.
  • the user can operate the application program installed on the user terminal 110.
  • the application program transmits user behavior data to the server 130 through the network 120, and the user terminal 110 can also receive data transmitted by the server 130 through the network 120.
  • the user terminal 110 may implement the photovoltaic module defect detection method provided by the embodiment of the present disclosure by running a program or thread, and transmit the obtained positional relationships of multiple three-dimensional spaces to the server 130 through the network 120.
  • the user terminal 110 may use its built-in application to execute a method for detecting defects in photovoltaic modules. In other examples, the user terminal 110 may execute the photovoltaic module defect detection method by calling an application stored externally of the user terminal 110.
  • the network 120 may be a single network, or a combination of at least two different networks.
  • the network 120 may include, but is not limited to, one or a combination of several of a local area network, a wide area network, a public network, and a private network.
  • the server 130 may be a single server or a server group, and each server in the group is connected through a wired or wireless network.
  • a server group can be centralized, such as a data center, or distributed.
  • the server 130 may be local or remote.
  • the database 140 can generally refer to a device having a storage function.
  • the database 140 is mainly used to store various data used, generated and output by the user terminal 110 and the server 130 in the work.
  • the database 140 stores location information, category information, input images, classifier training data sets, etc. of the salient area, and the server 130 reads the information or data required by the user from the database 140 and passes the information or data through
  • the network 120 is sent to the user terminal 110, and the user terminal 110 displays the positional relationship of multiple three-dimensional spaces, thereby facilitating user browsing.
  • the database 140 may be local or remote.
  • the database 140 may include various memories, such as random access memory (Random Access Memory, RAM), read-only memory (Read Only Memory, ROM), and so on.
  • RAM random access memory
  • ROM read-only memory
  • the storage devices mentioned above are just a few examples, and the storage devices that can be used by the system 100 are not limited thereto.
  • the database 140 may be connected or communicated with the server 130 or a part thereof via the network 120, or directly connected or communicated with the server 130, or a combination of the above two methods.
  • the database 140 may be a stand-alone device. In other examples, the database 140 may also be integrated in at least one of the user terminal 110 and the server 130. For example, the database 140 may be set on the user terminal 110 or on the server 130. For another example, the database 140 may also be distributed, a part of which is set on the user terminal 110 and the other part is set on the server 130.
  • the user terminal 110 performs processing and calculation based on the target image of the photovoltaic component to be inspected obtained by the image acquisition device 150, and obtains the position information of the salient area, and obtains the input image from the target image according to the position information ,
  • the input image is transmitted to the server 130 through the network 120, and is stored in the database 140.
  • the user terminal 110 also calls target detection and labeling tool software to label sample images and obtain training samples, which are transmitted to the server 130 via the network and stored in the database 140.
  • the user terminal 110 designs a classifier algorithm, calls a stored training sample, executes the classifier algorithm on the training sample to generate a classifier, and then inputs the stored input image into the classifier to output category information.
  • FIG. 16 is a schematic block diagram of a photovoltaic module defect detection device provided by an embodiment of the disclosure.
  • the photovoltaic module defect detection device 400 is, for example, suitable for implementing the photovoltaic module defect detection method provided by the embodiments of the present disclosure.
  • the photovoltaic module defect detection device 400 may be a terminal device or the like. It should be noted that the photovoltaic module defect detection device 400 shown in FIG. 16 is only an example, and not as a limitation to the embodiment of the present disclosure.
  • the photovoltaic module defect detection device 400 may include a processing device (such as a central processing unit, a graphics processor, etc.) 410, which may be based on a program stored in a read-only memory (ROM) 420 or from a storage device 480
  • the program loaded into the random access memory (RAM) 430 executes various appropriate actions and processing.
  • the RAM 430 also stores various programs and data required for the operation of the photovoltaic module defect detection device 400.
  • the processing device 410, the ROM 420, and the RAM 430 are connected to each other through a bus 440.
  • An input/output (I/O) interface 450 is also connected to the bus 440.
  • the following devices can be connected to the I/O interface 450: including input devices 460 such as touch screen, touch panel, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; including, for example, liquid crystal display (LCD), speakers, vibration An output device 470 such as a device; a storage device 480 such as a magnetic tape and a hard disk; and a communication device 490.
  • the communication device 490 may allow the photovoltaic module defect detection device 400 to communicate wirelessly or wiredly with other electronic devices to exchange data.
  • FIG. 16 shows a photovoltaic module defect detection device 400 including various devices, it should be understood that it is not required to implement or have all of the illustrated devices, and the photovoltaic module defect detection device 400 may alternatively be implemented or equipped More or fewer devices.
  • the photovoltaic module defect detection device 400 may further include a peripheral interface (not shown in the figure) and the like.
  • the peripheral interface can be various types of interfaces, such as a USB interface, a lightning interface, and the like.
  • the communication device 490 can communicate with a network and other devices through wireless communication, such as the Internet, an intranet, and/or a wireless network such as a cellular telephone network, a wireless local area network (LAN), and/or a metropolitan area network (MAN).
  • wireless communication such as the Internet, an intranet, and/or a wireless network such as a cellular telephone network, a wireless local area network (LAN), and/or a metropolitan area network (MAN).
  • Wireless communication can use any of a variety of communication standards, protocols and technologies, including but not limited to Global System for Mobile Communications (GSM), Enhanced Data GSM Environment (EDGE), Wideband Code Division Multiple Access (W-CDMA) , Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Bluetooth, Wi-Fi (e.g. based on IEEE 802.11a, IEEE 802.11b, IEEE 802.11g and/or IEEE 802.11n standards), voice transmission based on Internet protocol (VoIP), Wi-MAX, protocols used for e-mail, instant messaging and/or short message service (SMS), or any other suitable communication protocol.
  • GSM Global System for Mobile Communications
  • EDGE Enhanced Data GSM Environment
  • W-CDMA Wideband Code Division Multiple Access
  • CDMA Code Division Multiple Access
  • TDMA Time Division Multiple Access
  • Wi-Fi e.g. based on IEEE 802.11a, IEEE 802.11b, IEEE 802.11g and/or IEEE 802.11n standards
  • VoIP Internet protocol
  • Wi-MAX
  • the above-mentioned photovoltaic module defect detection method or classifier training method may be implemented as a computer software program.
  • an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer readable medium, and the computer program includes a method for performing the above-mentioned photovoltaic module defect detection method or classifier training method The program code.
  • the computer program may be downloaded and installed from the network through the communication device 490, or installed from the storage device 480, or installed from the ROM 420.
  • the functions defined in the photovoltaic module defect detection method or the classifier training method provided in the embodiments of the present disclosure can be executed.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)

Abstract

一种光伏组件缺陷的检测方法和装置、分类器的训练方法、终端设备以及非暂时性存储介质，该检测方法包括：对目标图像进行显著区域检测，获得至少一个显著区域的位置信息；根据该至少一个显著区域的位置信息获得至少一个输入图像；将该至少一个输入图像输入至分类器中，以由该分类器输出该至少一个显著区域的类别信息。该检测方法可以提高光伏组件缺陷的检测效率，实现实时检测。

Description

光伏组件缺陷的检测方法和装置、分类器的训练方法、终端设备及非暂时性存储介质
本申请要求于2019年5月13日递交的中国专利申请第201910395630.6的优先权，在此全文引用上述中国专利申请公开的内容以作为本申请的一部分。
技术领域
本公开实施例涉及一种光伏组件缺陷的检测方法和装置、分类器的训练方法、终端设备及非暂时性存储介质。
背景技术
目前，光伏发电已成为各种新能源中应用最为广泛的新能源发电形式。整个光伏电站中最核心的部件就是太阳能光伏电池板，其生产、安装过程中不可避免产生的缺陷将影响其工作效率，因此十分有必要对太阳能光伏电池板进行缺陷检测。
发明内容
本公开至少一实施例提供一种光伏组件缺陷的检测方法，包括：获取待检测光伏组件的目标图像；对所述目标图像进行显著区域检测，获得至少一个显著区域的位置信息；根据所述至少一个显著区域的位置信息获得至少一个输入图像；将所述至少一个输入图像输入至分类器中，以由所述分类器输出所述至少一个显著区域的类别信息。所述类别信息为所述至少一个显著区域所包括的缺陷的类别信息或正常区域的标识信息。
在一些示例中，所述目标图像为可见光图像。
在一些示例中，所述分类器经过训练，所述分类器的训练方法包括：将从样本图像中提取的局部图像作为训练样本输入至待训练的分类器中以由所述待训练的分类器输出所述局部图像的训练类别信息，所述局部图像包括所述样本图像中的一个局部区域，所述训练类别信息为所述局部图像所包括的 缺陷的类别信息或正常区域的标识信息；以及基于所述训练类别信息确定是否调整所述待训练的分类器。
在一些示例中，所述分类器的训练方法还包括：根据所述局部图像的位置信息从所述样本图像的相应位置提取所述局部图像。
在一些示例中，根据所述局部图像的位置信息从所述样本图像的相应位置提取所述局部图像包括：对所述样本图像进行标注得到包括所述局部区域的标注框并生成所述标注框的位置信息。
在一些示例中，根据所述局部图像的位置信息从所述样本图像的相应位置提取所述局部图像还包括：以所述标注框的中心点为中心、以所述标注框的长边为边长从所述样本图像中的对应位置提取正方形的所述局部区域作为所述局部图像。
在一些示例中，所述目标图像中的每个像素点的颜色信息包括RGB颜色信息；所述对所述目标图像进行显著区域检测，获得所述至少一个显著区域的位置信息，包括：对所述目标图像进行高斯平滑处理，得到新目标图像；将所述新目标图像中每个像素点的RGB颜色信息从当前的RGB色彩空间映射至CIELab色彩空间得到CIELab颜色信息；计算所述新目标图像中所有像素点的L、a和b颜色通道各自的通道平均值；针对每个颜色通道，计算所述新目标图像中每个像素点的颜色通道值与通道平均值的欧式距离，得到显著度图；对所述显著度图进行归一化，得到归一化的显著度图；对归一化的显著度图进行阈值分割，得到显著性二值图像；根据所述显著性二值图像中的至少一个连通区域的位置确定所述至少一个显著区域的位置信息。
在一些示例中，所述检测方法还包括：若所述类别信息为缺陷的类别信息，将所述显著区域的位置信息和类别信息用于对所述光伏组件的制作工艺进行改进。
在一些示例中，所述至少一个显著区域所包括的缺陷为所述光伏组件的外观缺陷。
在一些示例中，所述光伏组件的外观缺陷包括以下之一：污痕、划伤、脱晶、崩边、缺角、焊带偏移、碎片、穿孔、玻璃气泡。
本公开至少一实施例还提供一种分类器的训练方法，包括：将从样本图像中提取的局部图像作为训练样本输入至所述分类器中以由所述分类器输出 所述局部图像的训练类别信息，其中，所述局部图像包括所述样本图像中的一个局部区域；基于所述训练类别信息确定是否调整所述分类器。
在一些示例中，所述局部图像为包括单一缺陷的图像或包括正常区域的图像；所述训练类别信息为所述局部图像所包括的单一缺陷的类别信息或所述正常区域的标识信息。
在一些示例中，所述训练方法还包括：根据所述局部图像的位置信息从所述样本图像的相应位置提取所述局部图像。
在一些示例中，所述训练方法还包括：对所述样本图像进行处理以得到所述局部图像的位置信息。
在一些示例中，对所述样本图像进行处理以得到所述局部图像的信息包括：对所述样本图像进行标注得到包括所述局部区域的标注框并生成所述标注框的位置信息作为所述局部图像的位置信息。
在一些示例中，根据所述局部图像的位置信息从所述样本图像的相应位置提取所述局部图像包括：以所述标注框的中心点为中心、以所述标注框的长边为边长从所述样本图像中的对应位置提取正方形的所述局部区域作为所述局部图像。
本公开至少一些实施例还提供一种光伏组件缺陷的检测装置，包括：第一获取电路，用于获取待检测光伏组件的目标图像；第二获取电路，用于对所述目标图像进行显著区域检测，获得显著区域的位置信息；第三获取电路，用于根据所述显著区域的位置信息获得输入图像；分类电路，用于将所述输入图像输入至分类器中，以由所述分类器输出所述目标图像中显著区域的类别信息；所述类别信息为缺陷的类别信息或正常区域的标识信息。
本公开至少一些实施例还提供一种终端设备，包括处理器和存储器。所述存储器中存储有计算机程序指令，所述计算机程序指令当由所述处理器运行时，执行如上所述的检测方法或训练方法。
本公开至少一些实施例还提供一种非暂时性存储介质，非暂时性地存储有计算机程序指令，当所述计算机程序指令由计算机执行时可以实现上述检测方法或训练方法。
附图说明
为了更清楚地说明本发明实施例的技术方案，下面将对实施例的附图作简单地介绍，显而易见地，下面描述中的附图仅仅涉及本发明的一些实施例，而非对本发明的限制。
图1是根据本公开实施例示出的光伏组件缺陷的检测方法的流程图之一；
图2-图8是根据本公开实施例示出的光伏组件的外观缺陷的效果图；
图9是根据本公开实施例示出的光伏组件缺陷的检测方法的流程图之二；
图10是根据本公开实施例示出的光伏组件缺陷的检测方法的流程图之三；
图11是根据本公开实施例示出的光伏组件缺陷的检测方法的流程图之四；
图12是根据本公开实施例示出的一种光伏组件缺陷的检测装置的框图；
图13是根据本公开实施例示出的一种终端设备的结构示意图；
图14为本公开至少一实施例提供的一种存储介质的示意图；
图15为本公开至少一实施例提供的一种光伏组件缺陷的检测系统的示意图；以及
图16为本公开至少一实施例提供的一种光伏组件缺陷的检测装置的结构示意图。
具体实施方式
为使本公开实施例的目的、技术方案和优点更加清楚，下面将结合本公开实施例的附图，对本公开实施例的技术方案进行清楚、完整地描述。显然，所描述的实施例是本公开的一部分实施例，而不是全部的实施例。基于所描述的本公开的实施例，本领域普通技术人员在无需创造性劳动的前提下所获得的所有其他实施例，都属于本公开保护的范围。
除非另外定义，本公开使用的技术术语或者科学术语应当为本公开所属领域内具有一般技能的人士所理解的通常意义。本公开中使用的“第一”、“第二”以及类似的词语并不表示任何顺序、数量或者重要性，而只是用来区分不同的组成部分。同样，“一个”、“一”或者“该”等类似词语也 不表示数量限制，而是表示存在至少一个。“包括”或者“包含”等类似的词语意指出现该词前面的元件或者物件涵盖出现在该词后面列举的元件或者物件及其等同，而不排除其他元件或者物件。“连接”或者“相连”等类似的词语并非限定于物理的或者机械的连接，而是可以包括电性的连接，不管是直接的还是间接的。“上”、“下”、“左”、“右”等仅用于表示相对位置关系，当被描述对象的绝对位置改变后，则该相对位置关系也可能相应地改变。
发明人发现，在太阳能光伏组件外观检查时，仍主要依赖肉眼检测，对于大型组件中存在的细微缺陷，采用人工检查存在检查时间久和漏看缺陷点等问题，而实验室由于样品尺寸、型号以及实验流程导致的相关问题，无法采用类似工厂的缺陷判定机制，故使得检测效率相对较低，因此，如何提高光伏组件缺陷的检测效率是需要解决的一个问题。
图1是根据本公开实施例示出的一种光伏组件缺陷的检测方法。该光伏组件缺陷的检测方法可以应用于终端设备，该终端设备可以是个人计算机或服务器等。该光伏组件缺陷的检测方法，如图1所示，可以包括以下步骤101-104：
在步骤101中，获取待检测光伏组件的目标图像。例如，可以从图像获取装置获取待检测光伏组件的目标图像，又或者从其他设备(例如PC、服务器)或存储装置(例如，硬盘、U盘等)获取之前存储的待检测光伏组件的目标图像。例如，该图像获取装置可以包括工业相机，该工业相机对该光伏组件进行拍照以获取该目标图像。
在一个实施例中，当需要终端设备确定待检测光伏组件上是否存在外观缺陷、若存在外观缺陷则确定外观缺陷的类别时，可以获取待检测光伏组件的可见光图像，作为上述的目标图像。例如，待检测光伏组件的外观缺陷的类别可包括：
(1)外表面破碎、开裂、弯曲、不规整或损伤；
(2)光伏组件的有效工作区域的任何薄膜层有超过一个电池面积的10％以上的空隙、人眼看得见的腐蚀；例如，该光伏组件的有效工作区域为能够接收光并能够将该接收光转换为电信号的区域；
(3)在光伏组件的边缘和任何一部分电路之间形成连续的气泡或剥 层；
(4)丧失机械完整性。
例如，光伏组件典型的外观缺陷可如图2-图8所示。其中，图2所示的外观缺陷是电池片污痕/划伤，图3所示的外观缺陷是脱晶，图4所示的外观缺陷是崩边/缺角，图5所示的外观缺陷是焊带偏移，图6所示的外观缺陷是碎片/穿孔，图7所示的外观缺陷是玻璃气泡，图8所示的外观缺陷是玻璃划伤。
例如，该光伏组件的外观缺陷包括以下至少之一：污痕、划伤、脱晶、崩边、缺角、焊带偏移、碎片、穿孔、玻璃气泡。
在步骤102中，对所述目标图像进行显著区域检测，获得至少一个显著区域的位置信息。例如，该显著区域的位置信息为该显著区域在该目标图像中的位置信息。例如，该位置信息包括显著区域的一个顶点(例如左上角顶点)的坐标、以及该显著区域的长度(例如为最大长度)、宽度(例如为最大宽度)信息，从而可以根据该位置信息确定该显著区域的位置。例如，该显著区域为以该坐标确定的点为相应顶点、以该最大长度和最大宽度分别为长度和宽度的矩形区域。
在本公开实施例中，显著区域检测即为视觉显著性检测(Visual Saliency Detection)，是指通过智能算法模拟人的视觉特点，提取目标图像中的显著区域(人类感兴趣的区域)。在面对一个场景(例如该目标图像)时，人类自动地对感兴趣区域进行处理而选择性地忽略不感兴趣区域，这些人们感兴趣的区域被称之为显著区域。例如，该显著区域的颜色、亮度、边缘等特征与周围像素存在差异。在本实施例中，目标图像中的显著区域可以是目标图像中与大部分区域内容不同的区域。例如，待检测光伏组件的目标图像中，大部分区域是正常区域(不存在外观缺陷的区域)，小部分异常区域是可能存在外观缺陷的区域，则上述的异常区域即为显著区域。例如，由于相机的拍摄角度、光线等影响因素，目标图像中的显著区域也可以是正常区域，不存在外观缺陷。
在本实施例中，一个待检测光伏组件的目标图像中可能会存在一个、两个或两个以上的显著区域。因此，对目标图像进行显著区域检测，可以得到至少一个显著区域的位置信息，然后，根据至少一个显著区域的 位置信息可定位目标图像中的至少一个显著区域。
本公开实施例提供的光伏组件的缺陷检测方法通过先对目标图像进行显著区域检测，得到疑似缺陷的图像，这样有至少以下三方面优点：
一、将缺陷区域范围有效缩小，减小后续图像处理的数据量，从而降低了后面步骤中分类器的计算能力的要求并提高了计算速度；
二、通过进行显著区域检测，将目标图像的每个缺陷区域形成单个图像输入到分类器中，比起直接将整张目标图像输入到分类器或检测器，进一步提高了计算速度和准确度；
三、在分类器的训练过程中，由于训练样本也相应为样本图像中的单个缺陷区域的图像，比起输入整张样本图像进行训练，本公开实施例提供的检测方法可以有助于实现不同缺陷类别的样本数量的均衡，从而有助于提高分类器的综合检测能力，进而提高检测的准确性。
在一个实施例中，如图9所示，步骤102可包括以下步骤901-907：
在步骤901中，对所述目标图像进行高斯平滑处理，得到新目标图像。
在本实施例中，可以利用高斯函数对目标图像进行高斯平滑处理，得到新目标图像。这样，可以消除目标图像中的高斯噪声。其中，高斯平滑处理也可以称之为高斯滤波。通俗的讲，高斯滤波就是对整幅目标图像进行加权平均的过程。新目标图像中每一个像素点的值，都由其本身和邻域内的其他像素值经过加权平均后得到。
在本实施例中，例如，可以利用尺寸为5*5的高斯卷积核对目标图像进行高斯平滑处理，用高斯卷积核确定的邻域内像素的加权平均灰度值去替代高斯卷积核中心像素点的值，得到新目标图像。需要说明的是，高斯卷积核的尺寸可不限于5*5。
在本实施例中，例如，目标图像中的每个像素点的颜色信息可包括RGB颜色信息。例如，目标图像中的每个像素点的颜色信息可包括红色子像素值R、绿色子像素值G与蓝色子像素值B。在另一些示例中，目标图像中的每个像素点的颜色信息也可以包括YUV颜色信息，YUV颜色空间和RGB颜色空间是可以相互转换的。例如，可以先将YUV颜色空间转换为RGB颜色空间，然后再实施以下步骤。
在步骤902中，将所述新目标图像中每个像素点的RGB颜色信息从当前的RGB色彩空间映射至CIELab色彩空间得到CIELab颜色信息。
在本实施例中，可以先将所述新目标图像中每个像素点的RGB颜色信息从当前的RGB色彩空间映射至XYZ色彩空间得到XYZ信息，具体可利用下述的公式(1)，计算得到XYZ信息：
Figure PCTCN2020090032-appb-000001
其中，M为预设的矩阵，M的值可如下式(2)所示：
Figure PCTCN2020090032-appb-000002
然后，将新目标图像中每个像素点的XYZ信息转换至CIELab色彩空间得到CIELab颜色信息，具体可利用下述的公式(3)-(6)，计算得到：
L *＝116f(Y/Y n)-16  (3)
a*＝500[f(X/X n)-f(Y/Y n)]  (4)
b*＝200[f(Y/Y n)-f(Z/Z n)]  (5)
Figure PCTCN2020090032-appb-000003
一般情况下，Xn、Yn、Zn可都为1。X、Y、Z及t变量的取值范围可都是[0,1]，对应的L分量的取值范围为[0，100]，a和b分量的取值范围可都为[-127，127]。
在步骤903中，计算所述新目标图像中所有像素点的L、a和b颜色通道各自的通道平均值。
在本实施例中，可根据新目标图像中所有像素点的L颜色通道的颜 色通道值计算L颜色通道的通道平均值
Figure PCTCN2020090032-appb-000004
可根据新目标图像中所有像素点的a颜色通道的颜色通道值计算a颜色通道的通道平均值
Figure PCTCN2020090032-appb-000005
可根据新目标图像中所有像素点的b颜色通道的颜色通道值计算b颜色通道的通道平均值
Figure PCTCN2020090032-appb-000006
在步骤904中，针对每个颜色通道，计算所述新目标图像中每个像素点的颜色通道值与通道平均值的欧式距离，得到显著度图。
在本实施例中，针对L颜色通道，计算所述新目标图像中每个像素点的L颜色通道值与通道平均值
Figure PCTCN2020090032-appb-000007
的欧式距离，针对a颜色通道，计算所述新目标图像中每个像素点的a颜色通道值与通道平均值
Figure PCTCN2020090032-appb-000008
的欧式距离，针对b颜色通道，计算所述新目标图像中每个像素点的b颜色通道值与通道平均值
Figure PCTCN2020090032-appb-000009
的欧式距离，得到显著度图。
在步骤905中，对所述显著度图进行归一化，得到归一化的显著度图。
在步骤906中，对归一化的显著度图进行阈值分割，得到显著性二值图像。
在本实施例中，可以采用最大类间方差法(OTSU)对显著度图进行自适应阈值分割，显著性二值图像。需要说明的是，在实际应用时，也可以采用其他方法对归一化的显著度图进行阈值分割，得到显著性二值图像，不限于本公开实施例中提供的最大类间方差法(OTSU)。
在步骤907中，根据所述显著性二值图像中的至少一个连通区域的位置确定至少一个显著区域的位置信息。例如，该位置信息包括显著区域的一个顶点(例如左上角顶点)的坐标、以及该显著区域所涉及的最大长度和最大宽度，从而可以根据该位置信息确定该显著区域的位置。例如，该显著区域为以该坐标确定的点为相应顶点、以该最大长度和最大宽度分别为长度和宽度的矩形区域。
在本实施例中，所述显著性二值图像可存在一个、两个或两个以上的连通区域。每个连通区域的位置即为显著区域的位置。例如，该二值图像包括第一像素值区和第二像素值区，例如，该第一像素值区的像素值为1，为黑色区，对应正常区域；该第二像素值区的像素值为0，为白色区，对应显著区域，多个白色像素点彼此邻接连通所覆盖的区域被识别为显著区域。因此，可以根据显著性二值图像中的至少一个连通区域 的位置确定至少一个显著区域的位置信息。例如，可以将至少一个连通区域的位置信息对应地确定为至少一个显著区域的位置信息，也可以对至少一个连通区域的位置信息进行微调后，再将微调后的至少一个连通区域的位置信息确定为至少一个显著区域的位置信息，但不限于此。例如，在获得至少一个连通区域的位置信息后，可以将一部分或者每个连通区域的边缘向外扩展指定数目的像素，以获得增大的连通区域，然后将增大的连通区域的位置信息确定为显著区域的位置信息。其中，指定数目可以为1-3，但不限于此。这样，显著区域可以包括更多的信息，有利于提高缺陷识别的准确性。
在另一些示例中，还可以将所该目标图像输入至显著区域检测器中，以由该显著区域检测器输出该显著区域的位置信息。该显著区域检测器对该显著区域进行识别，并得到该显著区域的位置信息。例如，显著区域检测器可以通过适当方式实现，例如显著性检测神经网络；例如，可以利用深度学习技术对显著区域检测器进行训练。例如，一种算法可以参考Zhao T,Wu X.Pyramid Feature Selective Network for Saliency detection[C].2019IEEE Conference on Computer Vision and Pattern Recognition(CVPR).IEEE,2019.该算法利用基于上下文感知的金字塔注意网络生成显著图，其中金字塔网络基于VGG-16的conv3-3，conv4-3和conv5-3作为基础网络，采用不同扩展率(如3、5和7)的卷积，捕获多感受视野上下文信息。通过跨通道连接组合来自不同的atrous卷积层的特征映射和1*1维度减少特征，获得具有上下文感知信息的三种不同比例特征，并将两个较小的特征上采样到最大的一个，通过跨通道连接将他们组合为上下文感知金字塔网络的输出。使用上下文感知金字塔特征映射后的通道注意网络(CA)和低层次特征映射后的空间注意力网络(SA)提取高层次特征，然后融合CA和SA的输出得到显著图。显著性检测模型训练的损失函数为边缘保留损失函数，训练样本可利用软件IAT(Image Annotation Tool)进行标注。
在步骤103中，根据至少一个显著区域的位置信息获得至少一个输入图像。
在本实施例中，可以根据至少一个显著区域的位置信息从目标图像 中提取出至少一个显著区域，以获得至少一个输入图像，该输入图像即为该显著区域的图像。其中，针对每个显著区域，可以将从目标图像中提取出的显著区域作为输入图像。
在本实施例中，可通过如下方法从目标图像中提取出至少一个显著区域，以获得至少一个输入图像：针对每个显著区域对应的位置信息，确定对应的显著区域，针对每一个显著区域，以显著区域的外接矩形的长边为边长、以外接矩形的中心为中心提取正方形区域，得到输入图像。这样，可以使输入图像的形状或尺寸符合分类器对输入图像的要求。
在步骤104中，将所述至少一个输入图像输入至分类器中，以由所述分类器输出所述目标图像中至少一个显著区域的类别信息；所述类别信息为缺陷的类别信息或正常区域的标识信息(也即背景信息)。其中，正常区域中不存在缺陷。
在本实施例中，例如，该分类器可以是预先训练好的，可以预存在终端设备中或存储在终端设备的外部设备中，在需要时进行调用。
将至少一个输入图像输入至已训练的分类器中后，针对每个输入图像，由所述分类器识别出对应的显著区域的类别信息，当分类器输出显著区域的类别信息后，如果显著区域的类别信息为缺陷的类别信息，则可以根据显著区域的位置信息确定目标图像中缺陷区域的位置信息。当然，分类器若确定显著区域为正常区域时，也会输出正常区域的标识信息。终端设备在得到每个输入图像对应的显著区域的类别信息后，可以存储识别结果，并可以判断是否所有输入图像均识别完毕，若否，则控制分类器对下一个输入图像进行缺陷识别，若是，则输出所述目标图像中所有缺陷区域的位置信息和缺陷的类别信息。
在本实施例中，如果已训练的分类器可以识别两个缺陷类别，则可以采用二分类算法来识别缺陷的类别，如果已训练的分类器可以识别两个以上缺陷类别，则可以采用多分类算法来识别缺陷的类别，例如，可以识别输入图像中的缺陷类别是否是电池片污痕、崩边或玻璃气泡等。在一个实施例中，如图10所示，上述分类器的训练方法可以至少包括以下步骤1001-1002训练得到：
在步骤1001中，将从样本图像中提取的局部图像作为训练样本输入 至待训练的分类器中以由该待训练的分类器输出该局部图像的训练类别信息，其中，该局部图像包括所述样本图像中的一个局部区域，该训练类别信息为该局部图像所包括的缺陷的类别信息或正常区域的标识信息。
在步骤1002中，基于该训练类别信息确定是否调整该待训练的分类器。
在本实施例中，可以将训练样本输入至待训练的分类器中，对分类器进行训练，得到已训练的分类器。每个训练样本都有一个预设的与其对应的训练类别信息，训练样本的训练类别信息可包括缺陷的类别信息或负样本的标识信息。
例如，训练样本所包括的局部区域存在电池片污痕时，可以预设对应的训练类别信息为1；训练样本所包括的局部区域存在崩边时，可以预设对应的训练类别信息为2；训练样本所包括的局部区域存在玻璃气泡时，可以预设对应的训练类别信息为3；该训练样本所包括的局部区域为正常区域(不存在缺陷的区域)时，也即该训练样本为负样本时，可以预设对应的训练类别信息为4。
例如，如果输出的训练类别信息与该训练样本的预设的对应的训练类别信息不符，则确定调整该分类器(例如调整分类器的参数)直至该输出的训练类别信息与该预设的对应的训练类别信息相符，并进入下一个训练样本的训练。
例如，该局部图像为包括缺陷的包括单一缺陷的图像，也即仅包括一种缺陷，这样可以提高该缺陷的特征显著度，从而提高分类器的判断能力。由于在本公开实施例提供的检测方法中，是从目标图像中提取局部图像作为训练样本，因此可以有针对性地提取具有单一缺陷的区域，在训练样本的选取上更加灵活。
在一个实施例中，可以利用深度学习分类技术对分类器进行训练。例如，可以利用ResNet(Residual Neural Network，残差网络)对分类器进行训练。其中，ResNet的主要思想是在网络中增加了直连通道，即Highway Network(高速路神经网络)的思想。此前的网络结构是性能输入做一个非线性变换，而Highway Network则允许保留之前网络层的一 定比例的输出。ResNet的思想和Highway Network的思想也非常类似，允许原始输入信息直接传到后面的层中。
一般来讲，在ResNet网络结构中会用到两种残差模块，一种是以两个3*3的卷积网络串接在一起作为一个残差模块，另外一种是1*1、3*3、1*1的3个卷积网络串接在一起作为一个残差模块。
ResNet可以有不同的网络层数，ResNet的网络层数可以是50-layer(层)、101-layer或152-layer。不同网络层数的ResNet都是由上述的残差模块堆叠在一起实现的。
例如，该分类器的训练方法还可以包括：根据所述局部图像的位置信息从所述样本图像的相应位置提取所述局部图像作为该训练样本。
在一个实施例中，如图11所示，上述的训练样本可以通过以下步骤1101-1104得到：
在步骤1101中，对所述样本图像进行标注得到包括所述局部区域的标注框并生成所述标注框的位置信息作为该局部图像的位置信息。
在一个实施例中，可以对该样本图像进行人工标注，用标注框框出缺陷区域或正常区域，并生成该标注框的位置信息作为该局部图像的位置信息，同时可以添加该标注框对应的训练类别信息，用于标识标注框框出的区域的缺陷类别，或者标识负样本。例如，当标注框框出缺陷时，对应的训练类别信息可以为缺陷类别信息，当标注框框出的是正常区域不包括缺陷时，训练类别信息可以为负样本的标识信息。例如，该样本位置信息包括该标注框的一个顶点(例如左上角顶点)的坐标、以及该标注框的长度和宽度。
例如，当标注框框出的区域存在电池片污痕时，可以添加对应的训练类别信息为1；当标注框框出的区域存在崩边时，可以添加对应的训练类别信息为2；当标注框框出的区域存在玻璃气泡时，可以添加对应的训练类别信息为3；当标注框框出的区域为正常区域时，可以添加对应的训练类别信息为4。
例如，可以采用目标检测标注工具软件对该样本图像进行标注。该目标检测标注工具软件例如可以是LabelImage软件。
采用检测标注工具软件对图像进行标注具有准确度高等优点，可以 提高该样本图像所携带信息的准确性，从而获得高质量的训练样本，进而提高检测结果的准确度。
例如，可以选取不同位置或不同形貌的多个正常区域形成多个负样本图像对分类器进行训练，这样可以提高检测的准确率。
在步骤1102中，以该标注框的中心点为中心、以该标注框的长边为边长从所述样本图像中的对应位置提取正方形的局部区域作为该局部图像。例如，局部图像可以作为原始训练样本。
例如，原始的训练样本可以作为信息源，通过图像处理进行数据增强，得到多个衍生的训练样本(衍生训练样本)。
在步骤1103中，对所述训练样本进行数据增强，得到多个衍生的训练样本。
在本实施例中，可以对上述原始的训练样本进行旋转、缩放等图像处理实现数据增强，得到一系列的训练样本。当对上述原始的训练样本进行旋转时，可以依次对上述原始的训练样本旋转指定角度，得到一系列的衍生训练样本。当对上述原始的训练样本进行缩放时，可以依次对上述原始的训练样本进行缩小指定倍数，以及依次对上述原始的训练样本进行放大指定倍数，得到一系列的衍生训练样本。这样，可以增加训练样本的多样性，有利于提高分类器分类的准确度。该原始训练样本和对应的衍生训练样本具有相同的标签。
在本公开实施例中，可以将上述原始的训练样本以及通过数据增强得到的多个衍生的训练样本都输入至待训练的分类器中，对分类器进行训练，得到已训练的分类器。
在步骤1104中，根据所述训练类别信息生成训练样本的标签。该标签用于输入到分类器中。
在本实施例中，针对每个训练样本，可以根据训练样本的标注框对应的训练类别信息生成训练样本的标签。例如，当训练样本的标注框对应的训练类别信息为电池片污痕时，训练样本的标签为1，当训练样本的标注框对应的训练类别信息为崩边时，训练样本的标签为2，当训练样本的标注框对应的训练类别信息为玻璃气泡时，训练样本的标签为3；当训练样本为负样本时，训练样本的标签为4。
例如，该检测方法还可以包括：若步骤S104中输出的类别信息为缺陷的类别信息，将该显著区域的位置信息和类别信息用于对所述光伏组件的制作工艺进行改进。
若步骤S104中输出的类别信息为缺陷的类别信息，则表示该目标图像中存在缺陷，该显著区域的位置信息和类别信息即为该缺陷的位置信息和类别信息。本公开实施例提供的光伏组件的缺陷检测方法，通过获得缺陷的位置信息，可以对光伏组件上的缺陷进行定位，不仅满足质量检测的需求，还能够提供更多有效信息。
例如，如果该缺陷为污痕，且发生在该光伏组件的非有效工作区域，该缺陷不会对该光伏组件的正常工作造成影响，那么可以判断该缺陷属于可以容忍的缺陷。
例如，可以对缺陷的位置信息和类别信息进行分析，例如进行大数据分析。例如，在一些情形下，分析结果表明光伏组件的某个位置发生某种特定的缺陷的概率较高，可以考虑对于该缺陷形成的相关工艺步骤进行改进，从而有针对性地对工艺流程提供了反馈信息，有利于工艺流程的优化。
在本公开实施例中，由于先对待检测光伏组件的目标图像进行了显著区域检测，获得至少一个显著区域作为缺陷区域的候选区域，然后，根据至少一个显著区域对应的位置信息获得至少一个输入图像，并将上述的至少一个输入图像输入至分类器中，以由分类器输出目标图像中至少一个显著区域的类别信息，以便用户可以获知待检测光伏组件是否存在缺陷，当存在缺陷时缺陷的类别。其中，上述的类别信息为缺陷的类别信息或正常区域的标识信息。这样，可以减小图像处理的数据量，进而提高光伏组件缺陷的检测效率，实现实时检测。
本公开的至少一个实施例可以具有如下所述的至少一项有益效果：
(1)可以实现对光伏组件的外观缺陷的自动化检测，减少人力成本。
(2)考虑到光伏组件外观缺陷的种类繁多，而且每种外观缺陷的表观形态复杂，深度学习技术能满足算法精度要求。而且，考虑到实时自动化检测的工程性需求，本公开实施例将目标检测问题转为显著性区域检测和目标识别问题，这样既保证了算法精度又保证了算法的实时性。 本公开的实施例还提出了一种光伏组件缺陷的检测装置，如图12所示，该装置包括：
第一获取模块121，用于获取待检测光伏组件的目标图像；
第二获取模块122，用于对所述目标图像进行显著区域检测，获得至少一个显著区域对应的位置信息；
第三获取模块123，用于根据至少一个显著区域对应的位置信息获得至少一个输入图像；
分类模块124，用于将所述至少一个输入图像输入至分类器中，以由所述分类器输出所述目标图像中至少一个显著区域的类别信息；所述类别信息为缺陷的类别信息或正常区域的标识信息。
该第一获取模块121、第二获取模块122、第三获取模块123和第四分类模块124可以通过硬件(例如电路)模块、软件模块或它们的任意组合实现，本公开对此不作限制。
本公开的至少一个实施例中，由于先对待检测光伏组件的目标图像进行了显著区域检测，获得至少一个显著区域作为缺陷区域的候选区域，然后，根据至少一个显著区域对应的位置信息获得至少一个输入图像，并将上述的至少一个输入图像输入至分类器中，以由分类器输出目标图像中至少一个显著区域的类别信息，以便用户可以获知待检测光伏组件是否存在缺陷，当存在缺陷时缺陷的类别。其中，上述的类别信息为缺陷的类别信息或正常区域的标识信息。这样，可以减小图像处理的数据量，进而提高光伏组件缺陷的检测效率，实现实时检测。
本公开至少一实施例还提供一种分类器的训练方法，包括：将从样本图像中提取的局部图像作为训练样本输入至所述分类器中以由所述分类器输出所述局部图像的训练类别信息，其中，所述局部图像包括所述样本图像中的一个局部区域；以及基于所述训练类别信息确定是否调整所述分类器。
在本公开实施例提供的分类器的训练方法中，是从目标图像中提取局部图像作为训练样本，例如，比起直接将整张目标图像输入到分类器或检测器，具有以下至少三方面优点：
一、降低了单个训练样本的数据量，从而降低了分类器的计算能力 的要求并提高了计算速度；
二、训练样本的选取更加灵活；
三、有助于实现不同类别的样本数量的均衡，从而有助于提高分类器的综合检测能力，进而提高检测的准确性。
每个训练样本都有一个预设的对应的训练类别信息，例如，如果输出的训练类别信息与该训练样本的预设的对应的训练类别信息不符，则确定调整该分类器(例如调整分类器的参数)直至该输出的训练类别信息与该预设的对应的训练类别信息相符，并进入下一个训练样本的训练。
例如，该局部图像为包括单一缺陷的图像或包括正常区域的图像，所述训练类别信息为所述局部图像所包括的单一缺陷的类别信息或所述正常区域的标识信息。
例如，当该样本图像为光伏组件的图像，训练样本所包括的局部区域存在电池片污痕时，可以预设对应的训练类别信息为1；训练样本所包括的局部区域存在崩边时，可以预设对应的训练类别信息为2；训练样本所包括的局部区域存在玻璃气泡时，可以预设对应的训练类别信息为3；该训练样本所包括的局部区域为正常区域(不存在缺陷的区域)时，也即该训练样本为负样本时，可以预设对应的训练类别信息为4。
例如，该局部图像为包括缺陷的图像或包括正常区域的图像，所述训练类别信息为所述局部图像所包括的缺陷的类别信息或所正常区域的标识信息。
例如，该局部图像为包括缺陷的包括单一缺陷的图像，也即仅包括一种缺陷，这样可以提高该缺陷的特征显著度，从而提高分类器的判断能力。由于在本公开实施例提供的分类器的训练方法中，是从目标图像中提取局部图像作为训练样本，因此可以有针对性地提取具有单一缺陷的区域，在训练样本的选取上，比起将整张目标图像作为训练样本更加灵活。
例如，根据所述局部图像的位置信息从所述样本图像的相应位置提取所述局部图像。
例如，对所述样本图像进行处理以得到所述局部图像的位置信息。在一种示例中，可以对所述样本图像进行标注得到包括所述局部区域的 标注框并生成所述标注框的位置信息作为所述局部图像的位置信息；例如，该标注框的位置信息包括该标注框的一个顶点(例如左上角顶点)的坐标、以及该标注框的长度和宽度。
例如，以所述标注框的中心点为中心、以所述标注框的长边为边长从所述样本图像中的对应位置提取正方形的所述局部区域作为所述局部图像。
例如，可以采用目标检测标注工具软件对该样本图像进行标注。该目标检测标注工具软件例如可以是LabelImage软件。
例如，本公开至少一实施例提供的光伏组件缺陷的检测方法中使用的分类器可以经由上述分类器的训练方法进行训练得到。
图13是根据一示例性实施例示出的一种终端设备的框图。例如，终端设备1400可以被提供为一服务器，但不限于此。参照图13，设备1400包括处理组件1422，其进一步包括一个或多个处理器，以及由存储器1432所代表的存储器资源，用于存储可由处理部件1422的执行的指令，例如应用程序。存储器1432中存储的应用程序可以包括一个或一个以上的每一个对应于一组指令的模块。此外，处理组件1422被配置为执行指令，以执行上述用于光伏组件缺陷的检测方法。例如，处理组件1422与存储器1432通过总线系统连接。例如，总线系统可以是常用的串行、并行通信总线等，本公开的实施例对此不作限制。
设备1400还可以包括一个电源组件1426被配置为执行设备1400的电源管理，一个有线或无线网络接口1450被配置为将设备1400连接到网络，和一个输入输出(I/O)接口1458。设备1400可以操作基于存储在存储器1432的操作系统，例如Windows ServerTM，Mac OS XTM，UnixTM，LinuxTM，FreeBSDTM或类似。
例如，该处理组件1422可以是中央处理单元(CPU)、数字信号处理器(DSP)、图像处理器(GPU)或者具有数据处理能力和/或指令执行能力的其它形式的处理单元，可以为通用处理器或专用处理器，并且可以控制显示处理装置200中的其它组件以执行期望的功能。例如，该处理器可以为通用处理器或专用处理器，可以是基于X86或ARM架构的处理器等。
存储器1432可以包括一个或多个计算机程序产品，该计算机程序产品可以包括各种形式的计算机可读存储介质，例如易失性存储器和/或非易失性存储器。该易失性存储器例如可以包括随机存取存储器(RAM)和/或高速缓冲存储器(cache)等。该非易失性存储器例如可以包括只读存储器(ROM)、硬盘、闪存等。在计算机可读存储介质上可以存储一个或多个计算机程序指令，处理组件1422可以运行该程序指令，以实现本公开实施例中(由处理组件1422实现)的功能以及/或者其它期望的功能，例如光伏组件缺陷的检测方法等。在该计算机可读存储介质中还可以存储各种应用程序和各种数据，例如应用程序使用和/或产生的各种数据等。
在示例性实施例中，还提供了一种包括指令的非临时性计算机可读存储介质，如图14所示，存储介质600非暂时性地存储有计算机程序指令601。例如，当计算机程序指令601由计算机执行时可以执行根据上文所述的光伏组件缺陷的检测方法中的一个或多个步骤，或者执行上文所述的分类器的训练方法中的一个或多个步骤。
例如，该存储介质600可以为上述包括指令的存储器1432，上述指令可由设备1400的处理组件1422执行以完成上述检测方法或训练方法。例如，所述非临时性计算机可读存储介质可以是ROM、随机存取存储器(RAM)、CD-ROM、磁带、软盘和光数据存储设备等。
图15为一种可用于实施本公开实施例提供的光伏组件缺陷的检测方法的系统。如图14所示，该系统200可以包括用户终端110、网络120、服务器130、数据库140以及图像采集装置150。例如，该系统200可以用于实施本公开任一实施例提供的光伏组件缺陷的检测方法，其具体结构及功能等可以参考下文中用于实施光伏组件缺陷的检测方法的系统的相应内容。
用户终端110例如为电脑110-1或手机110-2。可以理解的是，用户终端110可以是能够执行数据处理的任何其他类型的电子设备，其可以包括但不限于台式电脑、笔记本电脑、平板电脑、智能手机、智能家居设备、可穿戴设备、车载电子设备、监控设备等。用户终端110也可以是设置有电子设备的任何装备，例如车辆、机器人等。
例如，图像采集装置150可以包括摄像头，该图像采集装置150对待检测的光伏组件进行拍照，将拍摄的光伏组件的图像数据通过网络120上传至服务器130，以供用户终端110调用，当然，也可以直接上传至用户终端110中。
用户可以对安装在用户终端110上的应用程序进行操作，应用程序通过网络120将用户行为数据传输给服务器130，用户终端110还可以通过网络120接收服务器130传输的数据。用户终端110可以通过运行程序或线程的方式实施本公开实施例提供的光伏组件缺陷的检测方法，并将得到的多个三维空间的位置关系通过网络120传输给服务器130。
在一些示例中，用户终端110可以利用其内置的应用程序执行光伏组件缺陷的检测方法。在另一些示例中，用户终端110可以通过调用用户终端110外部存储的应用程序执行光伏组件缺陷的检测方法。
网络120可以是单个网络，或至少两个不同网络的组合。例如，网络120可以包括但不限于局域网、广域网、公用网络、专用网络等中的一种或几种的组合。
服务器130可以是一个单独的服务器，或一个服务器群组，群组内的各个服务器通过有线的或无线的网络进行连接。一个服务器群组可以是集中式的，例如数据中心，也可以是分布式的。服务器130可以是本地的或远程的。
数据库140可以泛指具有存储功能的设备。数据库140主要用于存储用户终端110和服务器130在工作中所利用、产生和输出的各种数据。例如，数据库140中存储有显著区域的位置信息、类别信息、输入图像、分类器的训练数据集等，服务器130从数据库140中读取用户所需要的信息或数据，并将该信息或数据通过网络120发送至用户终端110，用户终端110显示多个三维空间的位置关系，从而便于用户浏览。数据库140可以是本地的或远程的。数据库140可以包括各种存储器、例如随机存取存储器(Random Access Memory，RAM)、只读存储器(Read Only Memory，ROM)等。以上提及的存储设备只是列举了一些例子，该系统100可以使用的存储设备并不局限于此。
数据库140可以经由网络120与服务器130或其一部分相互连接或 通信，或直接与服务器130相互连接或通信，或是上述两种方式的结合。
在一些示例中，数据库140可以是独立的设备。在另一些示例中，数据库140也可以集成在用户终端110和服务器130中的至少一个中。例如，数据库140可以设置在用户终端110上，也可以设置在服务器130上。又例如，数据库140也可以是分布式的，其一部分设置在用户终端110上，另一部分设置在服务器130上。
例如，在一些示例中，用户终端110基于图像采集装置150获取的待检测光伏组件的目标图像，进行处理和计算后，得到显著区域的位置信息，并根据该位置信息从目标图像中获得输入图像，该输入图像通过网络120传输给服务器130，并被保存至数据库140。例如，用户终端110还调用目标检测标注工具软件对样本图像进行标注并获得训练样本，该训练样本通过网络传输给服务器130并被保存至数据库140中。例如，用户终端110设计分类器算法，并调用储存的训练样本，在该训练样本上执行分类器算法生成分类器，然后将储存的输入图像输入到分类器中输出类别信息。
图16为本公开一实施例提供的光伏组件缺陷的检测装置的示意框图。该光伏组件缺陷的检测装置400例如适于用来实施本公开实施例提供的光伏组件缺陷的检测方法。光伏组件缺陷的检测装置400可以是终端设备等。需要注意的是，图16示出的光伏组件缺陷的检测装置400仅仅是一个示例，而不作为对本公开实施例的限制。
如图16所示，光伏组件缺陷的检测装置400可以包括处理装置(例如中央处理器、图形处理器等)410，其可以根据存储在只读存储器(ROM)420中的程序或者从存储装置480加载到随机访问存储器(RAM)430中的程序而执行各种适当的动作和处理。在RAM 430中，还存储有光伏组件缺陷的检测装置400操作所需的各种程序和数据。处理装置410、ROM 420以及RAM 430通过总线440彼此相连。输入/输出(I/O)接口450也连接至总线440。
通常，以下装置可以连接至I/O接口450：包括例如触摸屏、触摸板、键盘、鼠标、摄像头、麦克风、加速度计、陀螺仪等的输入装置460；包括例如液晶显示器(LCD)、扬声器、振动器等的输出装置470；包 括例如磁带、硬盘等的存储装置480；以及通信装置490。通信装置490可以允许光伏组件缺陷的检测装置400与其他电子设备进行无线或有线通信以交换数据。虽然图16示出了包括各种装置的光伏组件缺陷的检测装置400，但应理解的是，并不要求实施或具备所有示出的装置，光伏组件缺陷的检测装置400可以替代地实施或具备更多或更少的装置。
例如，该光伏组件缺陷的检测装置400还可以进一步包括外设接口(图中未示出)等。该外设接口可以为各种类型的接口，例如为USB接口、闪电(lighting)接口等。该通信装置490可以通过无线通信来与网络和其他设备进行通信，该网络例如为因特网、内部网和/或诸如蜂窝电话网络之类的无线网络、无线局域网(LAN)和/或城域网(MAN)。无线通信可以使用多种通信标准、协议和技术中的任何一种，包括但不局限于全球移动通信系统(GSM)、增强型数据GSM环境(EDGE)、宽带码分多址(W-CDMA)、码分多址(CDMA)、时分多址(TDMA)、蓝牙、Wi-Fi(例如基于IEEE 802.11a、IEEE 802.11b、IEEE 802.11g和/或IEEE 802.11n标准)、基于因特网协议的语音传输(VoIP)、Wi-MAX，用于电子邮件、即时消息传递和/或短消息服务(SMS)的协议，或任何其他合适的通信协议。
例如，根据本公开的实施例，上述光伏组件缺陷的检测方法或分类器的训练方法可以被实现为计算机软件程序。例如，本公开的实施例包括一种计算机程序产品，其包括承载在非暂态计算机可读介质上的计算机程序，该计算机程序包括用于执行上述光伏组件缺陷的检测方法或分类器的训练方法的程序代码。在这样的实施例中，该计算机程序可以通过通信装置490从网络上被下载和安装，或者从存储装置480安装，或者从ROM420安装。在该计算机程序被处理装置410执行时，可以执行本公开实施例提供的光伏组件缺陷的检测方法或分类器的训练方法中限定的功能。
以上所述仅是本发明的示范性实施方式，而非用于限制本发明的保护范围，本发明的保护范围由所附的权利要求确定。

Claims (20)

  1. 一种光伏组件缺陷的检测方法，包括：
    获取待检测光伏组件的目标图像；
    对所述目标图像进行显著区域检测，获得至少一个显著区域的位置信息；
    根据所述至少一个显著区域的位置信息获得至少一个输入图像；
    将所述至少一个输入图像输入至分类器中，以由所述分类器输出所述至少一个显著区域的类别信息，其中，所述类别信息为所述至少一个显著区域所包括的缺陷的类别信息或正常区域的标识信息。
  2. 根据权利要求1所述的检测方法，其中，所述目标图像为可见光图像。
  3. 根据权利要求1或2所述的检测方法，其中，所述分类器经过训练，所述分类器的训练方法包括：
    将从样本图像中提取的局部图像作为训练样本输入至待训练的分类器中以由所述待训练的分类器输出所述局部图像的训练类别信息，
    其中，所述局部图像包括所述样本图像中的一个局部区域，所述训练类别信息为所述局部图像所包括的缺陷的类别信息或正常区域的标识信息；
    基于所述训练类别信息确定是否调整所述待训练的分类器。
  4. 根据权利要求3所述的检测方法，其中，所述分类器的训练方法还包括：根据所述局部图像的位置信息从所述样本图像的相应位置提取所述局部图像。
  5. 根据权利要求4所述的检测方法，其中，根据所述局部图像的位置信息从所述样本图像的相应位置提取所述局部图像包括：
    对所述样本图像进行标注得到包括所述局部区域的标注框并生成所述标注框的位置信息作为所述局部图像的位置信息。
  6. 根据权利要求5所述的检测方法，其中，根据所述局部图像的位置信息从所述样本图像的相应位置提取所述局部图像还包括：
    以所述标注框的中心点为中心、以所述标注框的长边为边长从所述样本图像中的对应位置提取正方形的所述局部区域作为所述局部图像。
  7. 根据权利要求1-6任一所述的检测方法，其中，所述目标图像中的每 个像素点的颜色信息包括RGB颜色信息；
    所述对所述目标图像进行显著区域检测，获得所述至少一个显著区域的位置信息，包括：
    对所述目标图像进行高斯平滑处理，得到新目标图像；
    将所述新目标图像中每个像素点的RGB颜色信息从当前的RGB色彩空间映射至CIELab色彩空间得到CIELab颜色信息；
    计算所述新目标图像中所有像素点的L、a和b颜色通道各自的通道平均值；
    针对每个颜色通道，计算所述新目标图像中每个像素点的颜色通道值与通道平均值的欧式距离，得到显著度图；
    对所述显著度图进行归一化，得到归一化的显著度图；
    对归一化的显著度图进行阈值分割，得到显著性二值图像；
    根据所述显著性二值图像中的至少一个连通区域的位置确定所述至少一个显著区域的位置信息。
  8. 根据权利要求1-7任一所述的检测方法，其中，对所述目标图像进行显著区域检测，获得所述至少一个显著区域的位置信息，包括：
    将所述目标图像输入至显著区域检测器中，以由所述显著区域检测器输出所述至少一个显著区域的位置信息。
  9. 根据权利要求1-8任一所述的检测方法，还包括：
    若所述类别信息为缺陷的类别信息，将所述显著区域的位置信息和类别信息用于对所述光伏组件的制作工艺进行改进。
  10. 根据权利要求1-9任一所述的检测方法，其中，所述至少一个显著区域所包括的缺陷为所述光伏组件的外观缺陷。
  11. 根据权利要求10所述的检测方法，其中，所述光伏组件的外观缺陷包括以下之一：污痕、划伤、脱晶、崩边、缺角、焊带偏移、碎片、穿孔、玻璃气泡。
  12. 一种分类器的训练方法，包括：
    将从样本图像中提取的局部图像作为训练样本输入至所述分类器中以由所述分类器输出所述局部图像的训练类别信息，其中，所述局部图像包括所 述样本图像中的一个局部区域；
    基于所述训练类别信息确定是否调整所述分类器。
  13. 根据权利要求12所述的训练方法，其中，所述局部图像为包括单一缺陷的图像或包括正常区域的图像；
    所述训练类别信息为所述局部图像所包括的单一缺陷的类别信息或所述正常区域的标识信息。
  14. 根据权利要求12或13所述的训练方法，还包括：
    根据所述局部图像的位置信息从所述样本图像的相应位置提取所述局部图像。
  15. 根据权利要求14所述的训练方法，还包括：
    对所述样本图像进行处理以得到所述局部图像的位置信息。
  16. 根据权利要求15所述的训练方法，其中，对所述样本图像进行处理以得到所述局部图像的信息包括：
    对所述样本图像进行标注得到包括所述局部区域的标注框并生成所述标注框的位置信息作为所述局部图像的位置信息。
  17. 根据权利要求16所述的训练方法，其中，根据所述局部图像的位置信息从所述样本图像的相应位置提取所述局部图像包括：
    以所述标注框的中心点为中心、以所述标注框的长边为边长从所述样本图像中的对应位置提取正方形的所述局部区域作为所述局部图像。
  18. 一种光伏组件缺陷的检测装置，包括：
    第一获取电路，用于获取待检测光伏组件的目标图像；
    第二获取电路，用于对所述目标图像进行显著区域检测，获得至少一个显著区域的位置信息；
    第三获取电路，用于根据所述至少一个显著区域的位置信息获得至少一个输入图像；
    分类电路，用于将所述至少一个输入图像输入至分类器中，以由所述分类器输出所述至少一个显著区域的类别信息；所述类别信息为所述至少一个显著区域所包括的缺陷的类别信息或正常区域的标识信息。
  19. 一种终端设备，包括：
    处理器；
    存储器，其中，所述存储器中存储有计算机程序指令，所述计算机程序指令当由所述处理器运行时，执行如权利要求1-11任一所述的检测方法或如权利要求12-17中任一所述的训练方法。
  20. 一种非暂时性存储介质，非暂时性地存储有计算机程序指令，当所述计算机程序指令由计算机执行时可以实现如权利要求1-11任一所述的检测方法或如权利要求12-17中任一所述的训练方法。
PCT/CN2020/090032 2019-05-13 2020-05-13 光伏组件缺陷的检测方法和装置、分类器的训练方法、终端设备及非暂时性存储介质 WO2020228739A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910395630.6 2019-05-13
CN201910395630.6A CN110136110A (zh) 2019-05-13 2019-05-13 光伏组件缺陷的检测方法和装置

Publications (1)

Publication Number Publication Date
WO2020228739A1 true WO2020228739A1 (zh) 2020-11-19

Family

ID=67573703

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/090032 WO2020228739A1 (zh) 2019-05-13 2020-05-13 光伏组件缺陷的检测方法和装置、分类器的训练方法、终端设备及非暂时性存储介质

Country Status (2)

Country Link
CN (1) CN110136110A (zh)
WO (1) WO2020228739A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114841972A (zh) * 2022-05-09 2022-08-02 浙江大学 基于显著性图和语义嵌入特征金字塔的输电线路缺陷识别方法

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110136110A (zh) * 2019-05-13 2019-08-16 京东方科技集团股份有限公司 光伏组件缺陷的检测方法和装置
CN110738164B (zh) * 2019-10-12 2022-08-12 北京猎户星空科技有限公司 零件异常检测方法、模型训练方法及装置
CN111060520B (zh) * 2019-12-30 2021-10-29 歌尔股份有限公司 一种产品缺陷检测方法、装置与系统
CN111815560B (zh) * 2020-06-09 2024-04-05 理光软件研究所（北京）有限公司 光伏电站故障检测方法、装置、便携检测设备及存储介质
CN114972150B (zh) * 2021-02-24 2023-12-05 正泰集团研发中心(上海)有限公司 光伏组件尺寸缺陷检测方法
CN114581362B (zh) * 2021-07-22 2023-11-07 正泰集团研发中心(上海)有限公司 光伏组件缺陷检测方法、装置、电子设备和可读存储介质
CN114549393B (zh) * 2021-08-03 2024-01-19 正泰集团研发中心(上海)有限公司 图像标注方法、装置、设备及计算机存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101364263A (zh) * 2008-09-28 2009-02-11 腾讯科技（深圳）有限公司 对图像进行皮肤纹理检测的方法及系统
US20100150426A1 (en) * 2004-09-29 2010-06-17 Dainippon Screen Mfg. Co., Ltd. Apparatus and method for inspecting pattern
CN101996405A (zh) * 2010-08-30 2011-03-30 中国科学院计算技术研究所 一种快速的玻璃图像缺陷检测及分类方法及其装置
CN109613002A (zh) * 2018-11-21 2019-04-12 腾讯科技（深圳）有限公司 一种玻璃缺陷检测方法、装置和存储介质
CN109636772A (zh) * 2018-10-25 2019-04-16 同济大学 基于深度学习的不规则形状金属加工表面的缺陷检测方法
CN110136110A (zh) * 2019-05-13 2019-08-16 京东方科技集团股份有限公司 光伏组件缺陷的检测方法和装置

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5243785B2 (ja) * 2007-12-28 2013-07-24 日清紡ホールディングス株式会社 太陽電池検査装置及び太陽電池欠陥判定方法
EP2410319A1 (de) * 2010-07-23 2012-01-25 Solarpower GmbH Verfahren und System zur Detektion von defekten Solarmodulen
CN105046681A (zh) * 2015-05-14 2015-11-11 江南大学 一种基于SoC的图像显著性区域检测方法
CN104966285B (zh) * 2015-06-03 2018-01-19 北京工业大学 一种显著性区域的检测方法
CN105825238B (zh) * 2016-03-30 2019-04-30 江苏大学 一种视觉显著性目标的检测方法
CN106290388B (zh) * 2016-08-03 2018-09-28 国网山东省电力公司电力科学研究院 一种绝缘子故障自动检测方法
CN108631727B (zh) * 2018-03-26 2019-08-09 河北工业大学 一种基于卷积神经网络的太阳能电池板缺陷识别方法
CN108537239B (zh) * 2018-04-18 2020-11-17 西南科技大学 一种图像显著性目标检测的方法
CN108961154B (zh) * 2018-07-13 2022-12-23 福州大学 基于改进非下采样轮廓波变换的太阳电池热斑检测方法
CN109658376A (zh) * 2018-10-24 2019-04-19 哈尔滨工业大学 一种基于图像识别的表面缺陷识别方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100150426A1 (en) * 2004-09-29 2010-06-17 Dainippon Screen Mfg. Co., Ltd. Apparatus and method for inspecting pattern
CN101364263A (zh) * 2008-09-28 2009-02-11 腾讯科技（深圳）有限公司 对图像进行皮肤纹理检测的方法及系统
CN101996405A (zh) * 2010-08-30 2011-03-30 中国科学院计算技术研究所 一种快速的玻璃图像缺陷检测及分类方法及其装置
CN109636772A (zh) * 2018-10-25 2019-04-16 同济大学 基于深度学习的不规则形状金属加工表面的缺陷检测方法
CN109613002A (zh) * 2018-11-21 2019-04-12 腾讯科技（深圳）有限公司 一种玻璃缺陷检测方法、装置和存储介质
CN110136110A (zh) * 2019-05-13 2019-08-16 京东方科技集团股份有限公司 光伏组件缺陷的检测方法和装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114841972A (zh) * 2022-05-09 2022-08-02 浙江大学 基于显著性图和语义嵌入特征金字塔的输电线路缺陷识别方法

Also Published As

Publication number Publication date
CN110136110A (zh) 2019-08-16

Similar Documents

Publication Publication Date Title
WO2020228739A1 (zh) 光伏组件缺陷的检测方法和装置、分类器的训练方法、终端设备及非暂时性存储介质
WO2019233263A1 (zh) 视频处理方法、电子设备、计算机可读存储介质
WO2017088365A1 (zh) 一种肤色检测方法及装置
CN111612763A (zh) 手机屏幕缺陷检测方法、装置及系统、计算机设备及介质
WO2021082921A1 (zh) 电子设备背板外观瑕疵检测方法及设备
US8824747B2 (en) Skin-tone filtering
CN108509902B (zh) 一种驾驶员行车过程中手持电话通话行为检测方法
CN112241699A (zh) 物体缺陷类别识别方法、装置、计算机设备和存储介质
WO2020151299A1 (zh) 黄色禁停线识别方法、装置、计算机设备及存储介质
CN110827244A (zh) 一种用于电子设备外观瑕疵检测的方法与设备
CN108154149B (zh) 基于深度学习网络共享的车牌识别方法
CN111209907B (zh) 一种复杂光污染环境下产品特征图像人工智能识别方法
CN112132196B (zh) 一种结合深度学习和图像处理的烟盒缺陷识别方法
CN112966599B (zh) 关键点识别模型的训练方法、关键点识别方法及装置
WO2014169822A1 (zh) 图像分割方法和系统
WO2018068304A1 (zh) 一种图像匹配的方法及装置
CN111860316B (zh) 一种驾驶行为的识别方法、装置及存储介质
CN112989995B (zh) 文本检测方法、装置及电子设备
US20230065433A1 (en) Image processing method and apparatus, electronic device, and storage medium
WO2022002262A1 (zh) 基于计算机视觉的字符序列识别方法、装置、设备和介质
CN110866900A (zh) 水体颜色识别方法及装置
WO2022194079A1 (zh) 天空区域分割方法、装置、计算机设备和存储介质
CN116664586B (zh) 一种基于多模态特征融合的玻璃缺陷检测方法及系统
CN111461036A (zh) 一种利用背景建模增强数据的实时行人检测方法
US11270152B2 (en) Method and apparatus for image detection, patterning control method

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20805281

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20805281

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 20805281

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 20.07.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 20805281

Country of ref document: EP

Kind code of ref document: A1