CN116015192A - Control method of photovoltaic cleaning robot and photovoltaic cleaning robot - Google Patents
Control method of photovoltaic cleaning robot and photovoltaic cleaning robot Download PDFInfo
- Publication number
- CN116015192A CN116015192A CN202211586368.1A CN202211586368A CN116015192A CN 116015192 A CN116015192 A CN 116015192A CN 202211586368 A CN202211586368 A CN 202211586368A CN 116015192 A CN116015192 A CN 116015192A
- Authority
- CN
- China
- Prior art keywords
- photovoltaic
- bridge
- cleaning robot
- image data
- algorithm
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
Images
Classifications
-
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a control method of a photovoltaic cleaning robot and the photovoltaic cleaning robot, and the method comprises the following steps: collecting image data of a component of a current photovoltaic system in the walking process of the photovoltaic cleaning robot; detecting stain information on the surface of the photovoltaic module in each module image data in real time through a deep learning target detection algorithm; the stain information includes a stain location, a stain type; detecting the spot area on the surface of the photovoltaic module in each module image data in real time through a semantic segmentation algorithm; and controlling the photovoltaic cleaning robot to clean the stains on the surface of the photovoltaic module according to the stain information and the stain area on the surface of the photovoltaic module. The application can accurately wash in different modes according to the type and the area size of stains, so that the cleaning of the photovoltaic module is completed rapidly and efficiently, and the power generation rate of the photovoltaic module is improved.
Description
Technical Field
The invention relates to the field of cleaning of photovoltaic systems, in particular to a control method of a photovoltaic cleaning robot and the photovoltaic cleaning robot.
Background
Solar photovoltaic modules are the core part of a solar power generation system and are also the most important part of the solar power generation system. The solar energy is converted into electric energy, or is sent to a storage battery for storage, or the load is pushed to work.
The photovoltaic module is generally placed in the open air, but due to weather changes such as sand dust, strong wind and the like, and natural reasons such as bird droppings pollution and the like, covers such as stains, sand dust and the like which are difficult to remove are formed on the photovoltaic module, so that the power generation rate of the photovoltaic module is seriously affected.
Therefore, how to effectively remove the cover on the photovoltaic module is a urgent problem to be solved.
Disclosure of Invention
In order to solve the technical problems, the invention provides a control method of a photovoltaic cleaning robot and the photovoltaic cleaning robot.
Specifically, the technical scheme of the invention is as follows:
in one aspect, a control method of a photovoltaic cleaning robot includes:
collecting image data of a component of a current photovoltaic system in the walking process of the photovoltaic cleaning robot; the component image data comprises three or one of an infrared thermal imaging component diagram, a color component diagram and a black-and-white component diagram;
detecting stain information on the surface of the photovoltaic module in each module image data in real time through a deep learning target detection algorithm; the stain information includes a stain location, a stain type;
detecting the spot area on the surface of the photovoltaic module in each module image data in real time through a semantic segmentation algorithm; and controlling the photovoltaic cleaning robot to clean the stains on the surface of the photovoltaic module according to the stain information and the stain area on the surface of the photovoltaic module.
In one embodiment, further comprising: when the photovoltaic cleaning robot reaches the edge of a bridge of a photovoltaic system, acquiring bridge image data of the bridge of the photovoltaic system; the bridge image data comprise three or one of an infrared thermal imaging bridge diagram, a color bridge diagram and a black-and-white bridge diagram;
detecting whether a bridge of a photovoltaic system in each bridge image data is broken or not in real time through the deep learning target detection algorithm;
when the photovoltaic system bridge frame breaks, the photovoltaic cleaning robot is controlled to stop walking.
In one embodiment, after acquiring the bridge image data at the bridge of the photovoltaic system when the photovoltaic cleaning robot reaches the edge of the bridge of the photovoltaic system, the method further includes:
dividing the edge area of two adjacent photovoltaic modules in each bridge image data into an upper bridge edge area and a lower bridge edge area through a semantic segmentation algorithm;
the upper bridge edge area and the lower bridge edge area are matched into two lines through a traditional image algorithm, and a bridge included angle formed by the upper bridge edge area and the lower bridge edge area is calculated by calculating a line clamping angle formed by the two lines;
when the included angle of the bridge is larger than a preset included angle threshold, the photovoltaic cleaning robot cannot pass through the bridge of the photovoltaic system, and the photovoltaic cleaning robot is controlled to stop running.
In one embodiment, before the step of collecting the image data of the components of the current photovoltaic system during the walking process of the photovoltaic cleaning robot, the method further comprises:
when the photovoltaic cleaning robot is ready to start, acquiring environmental image data around the photovoltaic module; the environment image data comprises three or one of an infrared thermal imaging environment image, a color environment image and a black-and-white environment image;
carrying out weather type identification on each environmental image data through a classification algorithm in a deep learning target detection algorithm or a semantic segmentation algorithm;
when severe weather appears around the photovoltaic system, the photovoltaic cleaning robot is controlled to stop starting; the bad weather includes sand storm.
In one embodiment, the detection process of the deep learning target detection algorithm includes selecting length, width and position coordinates of a spot target area by a frame, and outputting the category and confidence of the spot in each spot target area;
the detection process of the semantic segmentation algorithm comprises the steps of segmenting out the outline of the stain target area and outputting the area size of the stain target area.
In one embodiment, after collecting the image data of the components of the current photovoltaic system during the walking process of the photovoltaic cleaning robot, the method further comprises:
detecting crack information on the surface of the photovoltaic module in each module image data in real time through a deep learning target detection algorithm, wherein the crack information comprises crack positions and crack types; and uploading the crack information to a monitoring platform.
In another aspect, a photovoltaic cleaning robot includes: the image acquisition unit is arranged on the free cradle head or the steering device and comprises an infrared thermal imaging camera, a color visible light camera and a black-white visible light camera, and is used for acquiring component image data of the current photovoltaic system in the walking process of the photovoltaic cleaning robot; the component image data comprises three or one of an infrared thermal imaging component diagram, a color component diagram and a black-and-white component diagram; the detection and identification unit is connected with the image acquisition unit and is provided with a deep learning target detection algorithm, and the detection and identification unit is used for detecting the spot information on the surface of the photovoltaic module in each module image data in real time through the deep learning target detection algorithm; the stain information includes a stain location, a stain type; the detection and identification unit is provided with a semantic segmentation algorithm and is also used for detecting the spot area on the surface of the photovoltaic module in each module image data in real time through the semantic segmentation algorithm; and the control unit is connected with the detection and identification unit and is used for controlling the photovoltaic cleaning robot to clean the stains on the surface of the photovoltaic module according to the stain information and the stain area on the surface of the photovoltaic module.
In one embodiment, the image acquisition unit is further configured to acquire bridge image data at a bridge of the photovoltaic system when the photovoltaic cleaning robot reaches an edge of the bridge of the photovoltaic system; the bridge image data comprise three or one of an infrared thermal imaging bridge diagram, a color bridge diagram and a black-and-white bridge diagram; the detection and identification unit is further used for detecting whether the bridge frame of the photovoltaic system in each bridge frame image data is broken or not in real time through the deep learning target detection algorithm; and the control unit is also used for controlling the photovoltaic cleaning robot to stop walking when the photovoltaic system bridge frame breaks.
In one embodiment, the detecting and identifying unit is further configured to divide an edge area of two adjacent photovoltaic modules in each bridge image data into an upper bridge edge area and a lower bridge edge area by using the semantic segmentation algorithm;
the detection and identification unit is provided with a traditional image algorithm and is also used for fitting the upper bridge frame edge area and the lower bridge frame edge area into two lines through the traditional image algorithm, and calculating a bridge included angle formed by the upper bridge frame edge area and the lower bridge frame edge area through calculating a wire clamping angle formed by the two lines;
and the control unit is further used for controlling the photovoltaic cleaning robot to stop running when the included angle of the bridge is larger than a preset included angle threshold value and the photovoltaic cleaning robot cannot pass through the photovoltaic system bridge.
In one embodiment, the image acquisition unit is further configured to acquire environmental image data around the photovoltaic module when the photovoltaic cleaning robot is ready to start; the environment image data comprises three or one of an infrared thermal imaging environment image, a color environment image and a black-and-white environment image;
the detection and identification unit is provided with a classification algorithm in a deep learning target detection algorithm or a semantic segmentation algorithm, and is also used for carrying out weather type identification on each environmental image data through the classification algorithm;
the control unit is also used for controlling the photovoltaic cleaning robot to stop starting when severe weather appears around the photovoltaic system; the bad weather includes sand storm.
In one embodiment, the detection process of the deep learning target detection algorithm includes selecting length, width and position coordinates of a spot target area by a frame, and outputting the category and confidence of the spot in each spot target area;
the detection process of the semantic segmentation algorithm comprises the steps of segmenting out the outline of the stain target area and outputting the area size of the stain target area.
In one embodiment, the detection and identification unit is further configured to detect crack information on the surface of the photovoltaic module in each module image data in real time through the deep learning target detection algorithm, where the crack information includes a crack position and a crack type; and uploading the crack information to a monitoring platform.
Compared with the prior art, the invention has at least one of the following beneficial effects:
the cleaning robot with visual function is used for detecting the position, size and type of the stains in real time through a visual system, a target detection algorithm and a semantic segmentation algorithm in the cleaning process; and then, according to the type and the area of the stains, performing accurate flushing in different modes, so that the cleaning of the photovoltaic module is rapidly and efficiently completed, and the power generation rate of the photovoltaic module is improved.
Drawings
The above features, technical features, advantages and implementation of the present invention will be further described in the following description of preferred embodiments with reference to the accompanying drawings in a clear and easily understood manner.
FIG. 1 is a flow chart of one embodiment of a method of controlling a photovoltaic cleaning robot provided herein;
FIG. 2 is a flow chart of another embodiment of a method of controlling a photovoltaic cleaning robot provided herein;
FIG. 3 is a flow chart of yet another embodiment of a method of controlling a photovoltaic cleaning robot provided herein;
FIG. 4 is a flow chart of yet another embodiment of a method of controlling a photovoltaic cleaning robot provided herein;
FIG. 5 is a flow chart of one embodiment of a method of controlling a photovoltaic cleaning robot provided herein;
fig. 6 is a block diagram of one embodiment of a photovoltaic cleaning robot provided herein.
Reference numerals illustrate:
10-image acquisition unit, 20-detection and identification unit, 30-control unit.
Detailed Description
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following description will explain the specific embodiments of the present invention with reference to the accompanying drawings. It is evident that the drawings in the following description are only examples of the invention, from which other drawings and other embodiments can be obtained by a person skilled in the art without inventive effort.
For simplicity of the drawing, only the parts relevant to the invention are schematically shown in each drawing, and they do not represent the actual structure thereof as a product. Additionally, in order to simplify the drawing for ease of understanding, components having the same structure or function in some of the drawings are shown schematically with only one of them, or only one of them is labeled. Herein, "a" means not only "only this one" but also "more than one" case.
It should be further understood that the term "and/or" as used in this specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.
In this context, it should be noted that the terms "mounted," "connected," and "connected" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected, unless explicitly stated or limited otherwise; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the above terms in the present invention will be understood in specific cases by those of ordinary skill in the art.
In addition, in the description of the present application, the terms "first," "second," and the like are used merely to distinguish between descriptions and are not to be construed as indicating or implying relative importance.
In one embodiment, referring to fig. 1 of the specification, the control method of a photovoltaic cleaning robot provided by the invention includes:
s100, acquiring image data of a component of a current photovoltaic system in the walking process of the photovoltaic cleaning robot; the component image data includes three or one of an infrared thermal imaging component diagram, a color component diagram and a black-and-white component diagram.
Specifically, an infrared thermal imaging camera, a color visible light camera and a black and white visible light camera are arranged on the photovoltaic cleaning robot. When the photovoltaic cleaning robot walks on the photovoltaic module to execute a cleaning task, an infrared thermal imaging camera of the photovoltaic cleaning robot acquires an infrared thermal imaging module diagram of the position where the photovoltaic cleaning robot is located; and/or a color visible light camera collects a color component diagram of the position where the photovoltaic cleaning robot is located; and/or a black-and-white visible light camera collects black-and-white component images of the position where the photovoltaic cleaning robot is located. Preferably, the photovoltaic cleaning robot collects three types of infrared thermal imaging component diagrams, color component diagrams and black-and-white component diagrams at the same time.
S110, detecting stain information on the surface of the photovoltaic module in each module image data in real time through a deep learning target detection algorithm; the stain information includes stain location and stain type.
Specifically, the infrared thermal imaging component diagram, the color component diagram and the black and white component diagram can be identified by a deep learning target detection algorithm, and whether the stains exist in the infrared thermal imaging component diagram, the color component diagram and the black and white component diagram, and the positions and the types of the stains can be identified. In addition, the deep learning target detection algorithm can be used for real-time detection only after training is completed.
The training process is (taking an infrared thermal imaging assembly diagram as an example for illustration): and collecting a large number of infrared thermal imaging component images, and uniformly dividing the infrared thermal imaging component images into three types, namely an infrared thermal imaging component sample image, an infrared thermal imaging component training image and an infrared thermal imaging component verification image. Marking the type of the stain in the infrared thermal imaging assembly sample graph, training the infrared thermal imaging assembly sample to generate a deep learning target detection algorithm, training the deep learning target detection algorithm by utilizing the infrared thermal imaging assembly training graph, and finally verifying the credibility of the deep learning target detection algorithm by utilizing the infrared thermal imaging assembly verification graph.
The detection process of the deep learning target detection algorithm comprises the steps of selecting length, width and position coordinates of a spot target area in a frame mode, and outputting the spot type, the spot position and the confidence of the spot in each spot target area.
S120, detecting the spot area on the surface of the photovoltaic module in each module image data in real time through a semantic segmentation algorithm.
Specifically, the areas of stains in the infrared thermal imaging component diagram, the color component diagram and the black-and-white component can be identified through a semantic segmentation algorithm, and the pixel level can be accurately achieved. The semantic segmentation algorithm is also required to be used for real-time detection after training is completed like a deep learning target detection algorithm.
The training process of the semantic segmentation algorithm comprises the following steps: and collecting a large number of infrared thermal imaging component images, and uniformly dividing the infrared thermal imaging component images into three types, namely an infrared thermal imaging component sample image, an infrared thermal imaging component training image and an infrared thermal imaging component verification image. Marking the type of the stain in the infrared thermal imaging assembly sample graph, training the infrared thermal imaging assembly sample to generate a deep learning target detection algorithm, training the deep learning target detection algorithm by utilizing the infrared thermal imaging assembly training graph, and finally verifying the credibility of the deep learning target detection algorithm by utilizing the infrared thermal imaging assembly verification graph.
The detection process of the semantic segmentation algorithm comprises the steps of segmenting out the outline of the stain target area and outputting the width, height and area size of the stains in the stain target area.
S130, controlling the photovoltaic cleaning robot to clean the stains on the surface of the photovoltaic module according to the stain information and the stain area on the surface of the photovoltaic module.
Specifically, when the spot on the photovoltaic module surface is stubborn spot and the spot area is great, photovoltaic cleaning robot washs with the repeated cleaning mode, and the reciprocal cleaning that is done in this position promptly, and the accessible airborne water jet equipment spouts after stubborn spot simultaneously, reciprocal cleaning. When the stains on the surface of the photovoltaic module are easy to clean, the photovoltaic cleaning robot cleans the stains in a normal cleaning mode.
In this embodiment, because the photovoltaic module is generally placed in the open air, due to sand and dust, bird droppings and the like, a covering such as dirt and dust which is difficult to remove is formed on the surface of the photovoltaic module, so that the power generation rate of the photovoltaic module is seriously affected. At present, through a cleaning robot with visual function, in the cleaning process, the position, the size of obstinate spot are detected through visual system, then carry out accurate washing to improve photovoltaic module's cleaning efficiency and generating rate.
In another embodiment, referring to fig. 2 of the specification, the control method of the photovoltaic cleaning robot provided by the invention further includes:
s200, when the photovoltaic cleaning robot reaches the edge of a bridge of the photovoltaic system, obtaining bridge image data of the bridge of the photovoltaic system; the bridge image data comprises three or one of an infrared thermal imaging bridge diagram, a color bridge diagram and a black-and-white bridge diagram.
Specifically, when the cleaning robot runs, the vision camera obtains the situation of the front Fang Guangfu component in real time in a video or photographing mode, and when the photovoltaic cleaning robot reaches the edge of a bridge of a photovoltaic system, the infrared thermal imaging camera of the photovoltaic cleaning robot acquires an infrared thermal imaging bridge diagram; and/or, the color visible light camera collects a color bridge diagram; and/or the black-and-white visible light camera collects black-and-white bridge diagram. Preferably, the photovoltaic cleaning robot collects three types of infrared thermal imaging bridge diagrams, color bridge diagrams and black-white bridge diagrams simultaneously. Preferably, the photovoltaic cleaning robot collects three types of infrared thermal imaging bridge diagrams, color bridge diagrams and black-white bridge diagrams simultaneously.
S210, detecting whether a bridge of the photovoltaic system in each bridge image data is broken or not in real time through a deep learning target detection algorithm.
Specifically, the deep learning target detection algorithm can detect whether the photovoltaic system bridge in the infrared thermal imaging bridge diagram, the color bridge diagram and the black-and-white bridge diagram is broken or not at the same time, and respectively output corresponding detection results. Detecting and judging whether the bridge frame of the photovoltaic system is broken or not from three dimensions, so that the detection accuracy is improved.
And S220, when the bridge frame of the photovoltaic system breaks, controlling the photovoltaic cleaning robot to stop walking.
Specifically, when breakage of the photovoltaic system bridge is detected in any one of the infrared thermal imaging bridge diagram, the color bridge diagram and the black-and-white bridge diagram, the photovoltaic cleaning robot is controlled to stop walking, and the photovoltaic cleaning robot is prevented from falling.
In this embodiment, since the photovoltaic modules are disposed and operated in the open air for a long time, the bridge at the junction between two adjacent photovoltaic systems may have a natural fracture or an unexpected fracture. According to the cleaning robot with the visual function, in the cleaning process, the edge of a photovoltaic system is reached, whether the bridge is broken or not can be detected in real time through a visual artificial intelligence algorithm, and the cleaning robot is controlled to stop advancing in advance, so that the falling of the cleaning robot is avoided.
In yet another embodiment, referring to fig. 3 of the specification, the present invention provides a control method for a photovoltaic cleaning robot:
s200, when the photovoltaic cleaning robot reaches the edge of a bridge of the photovoltaic system, obtaining bridge image data of the bridge of the photovoltaic system; the bridge image data comprises three or one of an infrared thermal imaging bridge diagram, a color bridge diagram and a black-and-white bridge diagram.
S211, dividing the edge area of two adjacent photovoltaic systems in each bridge image data into an upper bridge edge area and a lower bridge edge area through a semantic segmentation algorithm.
Specifically, two adjacent photovoltaic systems are connected through a bridge, and when the photovoltaic cleaning robot reaches the edge of the photovoltaic system, a bridge and a bridge image containing the edges of the two adjacent photovoltaic systems are obtained. And dividing the edge of the photovoltaic system where the photovoltaic cleaning robot is located in the bridge image into an upper bridge edge area by utilizing a semantic segmentation algorithm, and dividing the edge of the photovoltaic system opposite to the photovoltaic cleaning robot in the bridge image into a lower bridge edge area.
S221, fitting the upper bridge edge area and the lower bridge edge area into two lines through a traditional image algorithm, and calculating a bridge included angle formed by the upper bridge edge area and the lower bridge edge area through calculating a line clamping angle formed by the two lines.
Specifically, fitting an upper bridge edge area into one line through a traditional image algorithm, and fitting a lower bridge edge area into another line; the wire clamping angle of the two wire forming is the gradient of the bridge or the included angle between the two photovoltaic systems.
And S230, when the included angle of the bridge is larger than a preset included angle threshold, the photovoltaic cleaning robot cannot control the photovoltaic cleaning robot to stop running through the bridge of the photovoltaic group system.
Specifically, after calculating the abnormal inclination of the bridge or the included angle between the two photovoltaic systems, judging whether the inclination or the included angle exceeds a threshold value.
In the embodiment, when a bridge at the joint between the photovoltaic systems has a large span, whether the photovoltaic cleaning robot can cross can be detected in time through a visual artificial intelligence algorithm, and the cleaning robot is controlled to stop advancing in advance, so that the cleaning robot is prevented from falling down.
In still another embodiment, referring to fig. 4 of the specification, the control method of the photovoltaic cleaning robot provided by the invention further includes:
s300, when the photovoltaic cleaning robot is ready to start, acquiring environmental image data around a photovoltaic system; the environmental image data includes three or one of an infrared thermal imaging environmental map, a color environmental map, and a black-and-white environmental map.
Specifically, when the photovoltaic cleaning robot is ready to start, the image data of the front Fang Guangfu system and the surrounding conditions of the system are obtained in a form of photographing or video recording through a visual camera.
S310, performing weather type identification on each environmental image data through a classification algorithm in a deep learning target detection algorithm or a semantic segmentation algorithm.
Specifically, the classification algorithm may be a classification algorithm in the target detection algorithm, and may output a target class. The classification algorithm may be a classification algorithm in a semantic segmentation algorithm, classifying pixel by pixel. And identifying whether the current weather is rainy, snowy or sand storm by a classification algorithm in a target detection algorithm or a semantic segmentation algorithm.
S320, when severe weather appears around the photovoltaic system, the photovoltaic cleaning robot is controlled to stop starting; severe weather includes sand storms.
Specifically, if the current weather is sand storm or snowing, the photovoltaic cleaning robot stops going out to clean the photovoltaic module; and if the weather is rainy and sunny, the photovoltaic cleaning robot can go out to clean the photovoltaic module.
In the embodiment, whether the cleaning robot works is indicated by identifying weather conditions such as sand storm, rain, snow and the like through a visual artificial intelligence algorithm.
In one embodiment, referring to fig. 5 of the specification, the control method of the photovoltaic cleaning robot provided by the invention further includes:
s100, acquiring image data of a component of a current photovoltaic system in the walking process of the photovoltaic cleaning robot; the component image data includes three or one of an infrared thermal imaging component diagram, a color component diagram and a black-and-white component diagram.
S140, detecting crack information on the surface of the photovoltaic module in each module image data in real time through a deep learning target detection algorithm, wherein the crack information comprises crack positions and crack types; and uploading the crack information to a monitoring platform.
In the embodiment, whether hot spots, snail lines, glass cracks and the like appear on the solar panel or not is identified through a visual artificial intelligence algorithm, and the hot spots, snail lines, glass cracks and the like are sent to a monitoring platform. Meanwhile, the area of abnormal areas such as hot spots, snail lines, glass cracks and the like can be calculated through a semantic segmentation algorithm, and a report is sent to a client.
In one embodiment, referring to fig. 6 of the specification, the present invention provides a photovoltaic cleaning robot, including:
the image acquisition unit 10 is arranged on the free cradle head or the steering device, and the image acquisition unit 10 comprises an infrared thermal imaging camera, a color visible light camera and a black-and-white visible light camera and is used for acquiring component image data of the current photovoltaic system in the walking process of the photovoltaic cleaning robot; the component image data includes three or one of an infrared thermal imaging component diagram, a color component diagram and a black-and-white component diagram.
Specifically, a vision camera is arranged on the upper part of the photovoltaic cleaning robot, and the vision camera is provided with a cradle head or other steering devices. When the photovoltaic cleaning robot returns to the original route to run after cleaning of the single row is finished, the visual camera is rotated 180 degrees by rotating the cradle head or the steering device, and then the condition of the assembly can be detected in the return route. The vision camera comprises an infrared thermal imaging camera, a color visible light camera, a black and white visible light camera and a free cradle head or other steering devices.
The detection and identification unit 20 is connected with the image acquisition unit 10 and is provided with a deep learning target detection algorithm for detecting the stain information on the surface of the photovoltaic module in each module image data in real time through the deep learning target detection algorithm; the stain information includes stain location and stain type.
Specifically, the detection process of the deep learning target detection algorithm includes selecting length, width and position coordinates of a soil target area by a frame, and outputting the category and confidence of the soil in each soil target area. The deep learning target detection algorithm flow comprises the following steps: image collection- > data annotation & cleaning & transformation- > training/prediction- > output frame coordinates & confidence & category. As the three infrared thermal imaging component diagrams, the color component diagrams and the black-and-white component diagrams are collected, the target detection algorithm frame diagram can be predicted on three scales during prediction.
In this embodiment, the loss function in the deep learning target detection algorithm: firstly, replacing a box regression loss function by a GIoU (global information unit) and secondly, adding a focal loss to relieve the problem of class imbalance; classification loss and confidence loss binary cross entropy loss (BCE loss) is used.
BCE loss: calculating the loss, as follows;
wherein x is i Is the classified output of network prediction, y i (a value of magnitude between 0-1) a sigmoid transformation is performed on the classification output, i.e. an input is mapped between 0-1. y is i * Is a true category label, L class Is a classification loss value.
Block regression loss: the GIoU is replaced with CIoU as follows:
in the formula, v is the normalization of the aspect ratio difference value of the prediction frame and the real frame, and alpha is the normalization of the aspect ratio difference value of the prediction frame and the real frame; b refers to a predicted Box frame (a frame where an object in an image is detected by object detection, from which the center x, y of the object, the width w of the object, and the height h of the object can be known), B gt Frame representing true labeled target, B gt = (x_gt, y_gt, w_gt, h_gt), gt is an abbreviation for groundtrunk.
IoU represents the cross-over ratio, where b and b gt Represented by B and B gt P (·) is the euclidean distance and c is the diagonal length of the smallest closed box covering the two boxes (predicted and real).
Focal loss: the problem of sample class imbalance is relieved, alpha is class weight and used for balancing the problem of positive and negative sample imbalance, gamma represents sample weight difficult to divide and used for measuring sample difficult to divide and sample easy to divide.
FL(p t )＝-α(1-p t )γlog(p t )
In the above formula, the values of y are 1 and-1, which represent the foreground and the background respectively. The value range of p is 0-1, which is the probability of model prediction belonging to the prospect. Gamma is the modulation factor and is manually adjusted to be set in the range 0, 5. When γ is 0, it becomes the initial CE loss function; the weight factor alpha E [0,1]. When gamma is a positive sample, the weight factor is alpha; when negative, the weight factor is 1-alpha.
The detection and identification unit 20 is equipped with a semantic segmentation algorithm, and is further configured to detect, in real time, a soil area on the surface of the photovoltaic module in each module image data through the semantic segmentation algorithm.
Specifically, the detection process of the semantic segmentation algorithm includes segmenting out the outline of the stain target area, and outputting the width, height and area size of the stain target area. The semantic segmentation algorithm flow is as follows: image collection- > data annotation & cleaning & conversion- > training/prediction- > outputting the outline of the target.
The semantic segmentation algorithm frame diagram is firstly downsampled, and a plurality of skip connections are added for improving segmentation accuracy in the upsampling process.
In this embodiment, the loss function in the semantic segmentation algorithm: the image is pixel-level classified using a loss function that combines cross entropy loss and Dice loss. And the Lovasz-Softmax loss is added to divide the pictures of a plurality of categories, and the loss function can effectively improve the accuracy of dividing the small target. At training time, 60% of epochs are trained using cross entropy loss, 40% of epochs are finetung using lovassz loss;
the loss function of the combination of cross entropy and dice, the first half of the following is the cross entropy loss and the second half is DiceLoss:
wherein N is the total number of pixels of one picture or one batch number of pictures; y is Y b Is the prediction class result for the b-th pixel,
is the true class result for the b-th pixel.
Lovasz-softmax loss: the penalty function is a smooth extension of the discrete Jaccard (i.e., IOU) penalty;
wherein delta is Jc The convex closure representing the child Jaccard loss, the convex closure representing it being tight and the polynomial time being computable, C representing all classes, m (C) representing the Jaccard index and the vector errors of the C classes.
And the control unit 30 is connected with the detection and identification unit 20 and is used for controlling the photovoltaic cleaning robot to clean the stains on the surface of the photovoltaic module according to the stain information and the stain area on the surface of the photovoltaic module.
In this embodiment, the photovoltaic module is generally placed in the open air, and due to weather changes such as sand and dust, strong wind, etc., covers such as stains and sand and dust, etc., which are difficult to remove, are formed on the photovoltaic panel, which seriously affects the power generation rate. A cleaning robot with visual function is developed, the position and the size of stubborn stains are detected through a visual system, and then accurate flushing is performed.
In one embodiment, referring to fig. 1 of the specification, the present invention provides a photovoltaic cleaning robot, including:
the image acquisition unit 10 is further used for acquiring bridge image data at a bridge of the photovoltaic system when the photovoltaic cleaning robot reaches the edge of the bridge of the photovoltaic system; the bridge image data comprise three or one of an infrared thermal imaging bridge diagram, a color bridge diagram and a black-and-white bridge diagram;
the detection and identification unit 20 is further configured to detect whether a bridge of the photovoltaic system in each bridge image data is broken in real time through a deep learning target detection algorithm;
the control unit 30 is further configured to control the photovoltaic cleaning robot to stop traveling when the bridge of the photovoltaic system breaks.
In the embodiment, due to long-time work, when a bridge at the joint between the photovoltaic systems is broken or has a barrier or a large span in front, the bridge can be timely detected through a visual artificial intelligence algorithm, and the forward movement is stopped in advance to avoid the robot from falling.
In one embodiment, referring to fig. 1 of the specification, the present invention provides a photovoltaic cleaning robot, including:
the detection and identification unit 20 is further configured to divide an edge area of two adjacent photovoltaic systems in each bridge image data into an upper bridge edge area and a lower bridge edge area through a semantic segmentation algorithm;
the detection and identification unit 20 is provided with a traditional image algorithm, and is further used for combining the upper bridge edge area and the lower bridge edge area into two lines through the traditional image algorithm, and calculating a bridge included angle formed by the upper bridge edge area and the lower bridge edge area through calculating a line clamping angle formed by the two lines;
the control unit 30 is further configured to control the photovoltaic cleaning robot to stop running when the bridge angle is greater than the preset angle threshold value, and the photovoltaic cleaning robot cannot pass through the photovoltaic system bridge.
In the embodiment, due to long-time work, when a bridge at the joint between the photovoltaic systems is broken or has a barrier or a large span in front, the bridge can be timely detected through a visual artificial intelligence algorithm, and the forward movement is stopped in advance to avoid the robot from falling.
In one embodiment, referring to fig. 1 of the specification, the present invention provides a photovoltaic cleaning robot, including:
the image acquisition unit 10 is further used for acquiring environmental image data around the photovoltaic system when the photovoltaic cleaning robot is ready to start; the environment image data comprises three or one of an infrared thermal imaging environment image, a color environment image and a black-and-white environment image;
the detection and identification unit 20 is provided with a classification algorithm in a deep learning target detection algorithm or a semantic segmentation algorithm, and is further used for carrying out weather type identification on each environmental image data through the classification algorithm;
the control unit 30 is further configured to control the photovoltaic cleaning robot to stop starting when severe weather occurs around the photovoltaic system; severe weather includes sand storms.
In the embodiment, whether the cleaning robot works is indicated by identifying weather conditions such as sand storm, rain, snow and the like through a visual artificial intelligence algorithm.
In one embodiment, referring to fig. 1 of the specification, the present invention provides a photovoltaic cleaning robot, including:
detecting crack information on the surface of the photovoltaic module in each module image data in real time through a deep learning target detection algorithm, wherein the crack information comprises crack positions and crack types; and uploading the crack information to a monitoring platform.
In this embodiment, the crack types include cracks, hot spots, snail lines. When the cleaning robot runs, the vision camera obtains the condition of the photovoltaic module at the current position in real time in a video or photographing mode, the on-board controller identifies the crack condition of the surface of the photovoltaic module through a target detection algorithm, and then the module position where the crack is located is uploaded to a remote system; the operation and maintenance personnel can conveniently maintain regularly.
It should be noted that the above embodiments can be freely combined as needed. The foregoing is merely a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention, which are intended to be comprehended within the scope of the present invention.
Claims (10)
1. A control method of a photovoltaic cleaning robot, comprising:
collecting image data of a component of a current photovoltaic system in the walking process of the photovoltaic cleaning robot; the component image data comprises three or one of an infrared thermal imaging component diagram, a color component diagram and a black-and-white component diagram;
detecting stain information on the surface of the photovoltaic module in each module image data in real time through a deep learning target detection algorithm; the stain information includes a stain location, a stain type;
detecting the spot area on the surface of the photovoltaic module in each module image data in real time through a semantic segmentation algorithm;
and controlling the photovoltaic cleaning robot to clean the stains on the surface of the photovoltaic module according to the stain information and the stain area on the surface of the photovoltaic module.
2. The method for controlling a photovoltaic cleaning robot according to claim 1, further comprising:
when the photovoltaic cleaning robot reaches the edge of a bridge of a photovoltaic system, acquiring bridge image data of the bridge of the photovoltaic system; the bridge image data comprise three or one of an infrared thermal imaging bridge diagram, a color bridge diagram and a black-and-white bridge diagram;
detecting whether a bridge of a photovoltaic system in each bridge image data is broken or not in real time through the deep learning target detection algorithm;
when the photovoltaic system bridge frame breaks, the photovoltaic cleaning robot is controlled to stop walking.
3. The method for controlling a photovoltaic cleaning robot according to claim 2, wherein when the photovoltaic cleaning robot reaches the edge of a bridge of a photovoltaic system, after acquiring the bridge image data at the bridge of the photovoltaic system, the method further comprises:
dividing the edge areas of two adjacent photovoltaic systems in each bridge image data into an upper bridge edge area and a lower bridge edge area through a semantic segmentation algorithm;
the upper bridge edge area and the lower bridge edge area are matched into two lines through a traditional image algorithm, and a bridge included angle formed by the upper bridge edge area and the lower bridge edge area is calculated by calculating a line clamping angle formed by the two lines;
when the included angle of the bridge is larger than a preset included angle threshold, the photovoltaic cleaning robot cannot pass through the bridge of the photovoltaic system, and the photovoltaic cleaning robot is controlled to stop running.
4. The method for controlling a photovoltaic cleaning robot according to claim 1, further comprising, before the step of collecting the image data of the components of the current photovoltaic system during the traveling of the photovoltaic cleaning robot:
when the photovoltaic cleaning robot is ready to start, acquiring environmental image data around the photovoltaic module; the environment image data comprises three or one of an infrared thermal imaging environment image, a color environment image and a black-and-white environment image;
carrying out weather type identification on each environmental image data through a classification algorithm in a deep learning target detection algorithm or a semantic segmentation algorithm;
when severe weather appears around the photovoltaic system, the photovoltaic cleaning robot is controlled to stop starting; the bad weather includes sand storm.
5. The control method of a photovoltaic cleaning robot according to any one of claims 1 to 4, characterized by: after collecting the image data of the components of the current photovoltaic system in the walking process of the photovoltaic cleaning robot, the method further comprises the following steps:
detecting crack information on the surface of the photovoltaic module in each module image data in real time through a deep learning target detection algorithm, wherein the crack information comprises crack positions and crack types; and uploading the crack information to a monitoring platform.
6. A photovoltaic cleaning robot, comprising:
the image acquisition unit is arranged on the free cradle head or the steering device and comprises an infrared thermal imaging camera, a color visible light camera and a black-white visible light camera, and is used for acquiring component image data of the current photovoltaic system in the walking process of the photovoltaic cleaning robot; the component image data comprises three or one of an infrared thermal imaging component diagram, a color component diagram and a black-and-white component diagram;
the detection and identification unit is connected with the image acquisition unit and is provided with a deep learning target detection algorithm, and the detection and identification unit is used for detecting the spot information on the surface of the photovoltaic module in each module image data in real time through the deep learning target detection algorithm; the stain information includes a stain location, a stain type;
the detection and identification unit is provided with a semantic segmentation algorithm and is also used for detecting the spot area on the surface of the photovoltaic module in each module image data in real time through the semantic segmentation algorithm;
and the control unit is connected with the detection and identification unit and is used for controlling the photovoltaic cleaning robot to clean the stains on the surface of the photovoltaic module according to the stain information and the stain area on the surface of the photovoltaic module.
7. The photovoltaic cleaning robot of claim 6, wherein:
the image acquisition unit is further used for acquiring bridge image data of a bridge of the photovoltaic system when the photovoltaic cleaning robot reaches the edge of the bridge of the photovoltaic system; the bridge image data comprise three or one of an infrared thermal imaging bridge diagram, a color bridge diagram and a black-and-white bridge diagram;
the detection and identification unit is further used for detecting whether the bridge frame of the photovoltaic system in each bridge frame image data is broken or not in real time through the deep learning target detection algorithm;
and the control unit is also used for controlling the photovoltaic cleaning robot to stop walking when the photovoltaic system bridge frame breaks.
8. The photovoltaic cleaning robot of claim 7, wherein:
the detection and identification unit is further used for dividing the edge area of two adjacent photovoltaic modules in each bridge image data into an upper bridge edge area and a lower bridge edge area through the semantic segmentation algorithm;
the detection and identification unit is provided with a traditional image algorithm and is also used for fitting the upper bridge frame edge area and the lower bridge frame edge area into two lines through the traditional image algorithm, and calculating a bridge included angle formed by the upper bridge frame edge area and the lower bridge frame edge area through calculating a wire clamping angle formed by the two lines;
and the control unit is further used for controlling the photovoltaic cleaning robot to stop running when the included angle of the bridge is larger than a preset included angle threshold value and the photovoltaic cleaning robot cannot pass through the photovoltaic system bridge.
9. The photovoltaic cleaning robot of claim 6, wherein:
the image acquisition unit is further used for acquiring environment image data around the photovoltaic system when the photovoltaic cleaning robot is ready to start; the environment image data comprises three or one of an infrared thermal imaging environment image, a color environment image and a black-and-white environment image;
the detection and identification unit is provided with a classification algorithm in a deep learning target detection algorithm or a semantic segmentation algorithm, and is also used for carrying out weather type identification on each environmental image data through the classification algorithm;
the control unit is also used for controlling the photovoltaic cleaning robot to stop starting when severe weather appears around the photovoltaic system; the bad weather includes sand storm.
10. A photovoltaic cleaning robot according to any one of claims 6 to 9, characterized in that:
the detection and identification unit is further used for detecting crack information on the surface of the photovoltaic module in each module image data in real time through the deep learning target detection algorithm, wherein the crack information comprises crack positions and crack types; and uploading the crack information to a monitoring platform.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211586368.1A CN116015192A (en) | 2022-12-09 | 2022-12-09 | Control method of photovoltaic cleaning robot and photovoltaic cleaning robot |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211586368.1A CN116015192A (en) | 2022-12-09 | 2022-12-09 | Control method of photovoltaic cleaning robot and photovoltaic cleaning robot |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116015192A true CN116015192A (en) | 2023-04-25 |
Family
ID=86032577
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211586368.1A Pending CN116015192A (en) | 2022-12-09 | 2022-12-09 | Control method of photovoltaic cleaning robot and photovoltaic cleaning robot |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116015192A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116346022A (en) * | 2023-05-31 | 2023-06-27 | 威驰腾(福建)汽车有限公司 | Roof solar device and vehicle |
-
2022
- 2022-12-09 CN CN202211586368.1A patent/CN116015192A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116346022A (en) * | 2023-05-31 | 2023-06-27 | 威驰腾(福建)汽车有限公司 | Roof solar device and vehicle |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110197203B (en) | Bridge pavement crack classification and identification method based on width learning neural network | |
| Di Tommaso et al. | A multi-stage model based on YOLOv3 for defect detection in PV panels based on IR and visible imaging by unmanned aerial vehicle | |
| CN113362285B (en) | Steel rail surface damage fine-grained image classification and detection method | |
| CN108108772B (en) | Insulator pollution flashover state detection method based on aerial image of distribution line | |
| CN110060508B (en) | Automatic ship detection method for inland river bridge area | |
| Wang et al. | Intelligent monitoring of photovoltaic panels based on infrared detection | |
| de Oliveira et al. | Automatic fault detection of photovoltaic array by convolutional neural networks during aerial infrared thermography | |
| CN115909093A (en) | Power equipment fault detection method based on unmanned aerial vehicle inspection and infrared image semantic segmentation | |
| Hascoet et al. | Fasterrcnn monitoring of road damages: Competition and deployment | |
| CN116015192A (en) | Control method of photovoltaic cleaning robot and photovoltaic cleaning robot | |
| Prabhakaran et al. | Deep Learning-Based Model for Defect Detection and Localization on Photovoltaic Panels. | |
| CN110676753B (en) | Intelligent inspection robot for power transmission line | |
| CN112446246A (en) | Image occlusion detection method and vehicle-mounted terminal | |
| CN112329584A (en) | Method, system and equipment for automatically identifying foreign matters in power grid based on machine vision | |
| CN114723675A (en) | Photovoltaic module detection method, device, equipment and storage medium | |
| Montanez et al. | Photovoltaic module segmentation and thermal analysis tool from thermal images | |
| CN113792638A (en) | Thermal power plant rain drainage port pollutant identification method based on parallelgram-Yolov 4 | |
| CN112183403A (en) | Photovoltaic cleaning robot cleaning speed adjusting method based on computer vision | |
| CN108198164A (en) | A kind of detection device and method of cable tunnel cable integrity | |
| CN117036825A (en) | Solar cell panel detection method, medium and system | |
| CN116310274A (en) | State evaluation method for power transmission and transformation equipment | |
| CN115690807A (en) | Box number identification method based on OCR technology | |
| CN115909245A (en) | Visual multi-task processing method based on deep learning | |
| Zhou et al. | Water photovoltaic plant contaminant identification using visible light images | |
| Huang et al. | Study on a boat-assisted drone inspection scheme for the modern large-scale offshore wind farm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |