WO2020181817A1 - 一种基于神经网络模型的植物图像的识别方法及装置 - Google Patents
一种基于神经网络模型的植物图像的识别方法及装置 Download PDFInfo
- Publication number
- WO2020181817A1 WO2020181817A1 PCT/CN2019/118395 CN2019118395W WO2020181817A1 WO 2020181817 A1 WO2020181817 A1 WO 2020181817A1 CN 2019118395 W CN2019118395 W CN 2019118395W WO 2020181817 A1 WO2020181817 A1 WO 2020181817A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- plant
- preset
- image
- resolution
- sub
- 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.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
- G06F18/2193—Validation; Performance evaluation; Active pattern learning techniques based on specific statistical tests
Definitions
- This application relates to the field of image recognition technology, in particular to a method and device, storage medium, and computer equipment for recognizing plant images based on neural network models.
- Atractylodes which belongs to the genus Atractylodes of the Atractylaceae, and the two plants have different medicinal effects, but the appearance is very similar.
- atractylodes japonicus affect the efficacy of atractylodes japonicus.
- the present application provides a method and device, storage medium, and computer equipment for recognizing plant images based on a neural network model, which helps to improve the recognition efficiency of plant species.
- a method for recognizing plant images based on a neural network model including:
- the plant sample sub-images of the preset resolution are respectively sampled with replacement to obtain a training sample set and a test sample set, wherein the first preset The number of times is greater than the second preset number of times;
- the plant image to be recognized is received, and the type of the plant image to be recognized is recognized by using the trained model.
- a plant image recognition device based on a neural network model including:
- the sample image interception module is used to perform image interception on the plant sample image according to preset plant parts to obtain plant sample sub-images corresponding to the preset plant parts;
- a sample resolution processing module configured to perform resolution preprocessing on the plant sample sub-image to obtain a plant sample image with a preset resolution
- the sample set establishment module is used to perform sampling with replacement on the plant sample sub-images of the preset resolution according to the first preset number and the second preset number, respectively, to obtain a training sample set and a test sample set, wherein , The first preset number of times is greater than the second preset number of times;
- the model training module is used to train the MobileNetV2 lightweight convolutional neural network model using the training sample set until the success rate of the trained model recognizing the test sample set is greater than or equal to the preset success rate;
- the image recognition module is used to receive the plant image to be recognized, and use the trained model to recognize the type of the plant image to be recognized.
- a storage medium on which a computer program is stored, and the program is executed by a processor to realize the above-mentioned method for recognizing plant images based on a neural network model.
- a computer device including a storage medium, a processor, and a computer program stored on the storage medium and running on the processor.
- the processor executes the program, the above-mentioned neural-based Recognition method of plant image based on network model.
- the present application provides a method and device, storage medium, and computer equipment for recognizing plant images based on neural network models. After the plant sample sub-images corresponding to the preset plant parts are intercepted, the intercepted plant images The sample sub-images are subjected to resolution preprocessing, and the plant sample sub-images after the resolution preprocessing are used to establish a training sample set and a test sample set, thereby using the training sample set to train the corresponding MobileNetV2 convolutional neural network model, so that after training The recognition success rate of the test sample set of the model reaches the preset success rate standard, so that the new plant image to be recognized can be classified and the plant species can be recognized through the trained model. Compared with the prior art relying on experienced technicians to manually classify plants, this application can use images of plants to be identified taken by mobile devices equipped with trained models to identify the types of plants through the model, which improves the The efficiency of type recognition saves labor costs.
- FIG. 1 shows a schematic flowchart of a method for recognizing plant images based on a neural network model provided by an embodiment of the present application
- FIG. 2 shows a schematic flowchart of another method for recognizing plant images based on a neural network model provided by an embodiment of the present application
- FIG. 3 shows a schematic structural diagram of a plant image recognition device based on a neural network model provided by an embodiment of the present application
- FIG. 4 shows a schematic structural diagram of another plant image recognition device based on a neural network model provided by an embodiment of the present application
- Fig. 5 shows a schematic diagram of the physical structure of a computer device provided by an embodiment of the present application.
- a method for recognizing plant images based on a neural network model includes:
- Step 101 Perform image interception on a plant sample image according to a preset plant part to obtain a plant sample sub-image corresponding to the preset plant part.
- Atractylodes Rhizome are similar to those of Atractylodes Rhizome, but the root features are different.
- For sample images of Atractylodes Rhizome and Atractylodes Rhizome Use root features for training.
- the intercepted image includes a part of the original plant sample image. The key features of the plant are selected and the secondary features are ignored, so that the intercepted image is used for training to avoid the loss of the key feature information of the image and speed up the training speed and improve the modeling effectiveness.
- Step 102 Perform resolution preprocessing on the plant sample sub-image to obtain a plant sample image with a preset resolution.
- the plant sample image is intercepted to obtain the plant sample sub-image
- the size of is unified, specifically, all plant sample sub-images can be processed into images of preset resolution.
- Step 103 According to the first preset times and the second preset times, randomly sample the plant sample sub-images of the preset resolution with replacement to obtain the training sample set and the test sample set, where the first preset The number of times is greater than the second preset number of times.
- the Bootstrap method is used to establish a training sample set and a test sample set, and the sub-images of the plant sample after resolution processing are sampled with replacement according to the first preset number of times and the second preset number of times, respectively, to establish the training sample Sets and test sample sets, where the training sample set and the test sample set both contain multiple sub-images of the Atractylodes Rhizome sample and the sub-images of the Atractylodes Rhizome sample with the Atractylodes label, and generally speaking the number of samples in the training sample set
- the number of samples is higher than the test sample set, that is, the first preset number of times is greater than the second preset number of times.
- the sub-image of the plant sample can be sampled 2000 times with replacement to create a training sample set, which is performed 300 times Sampling with replacement establishes a test sample set.
- sampling can be random sampling or stratified sampling.
- Step 104 Use the training sample set to train the MobileNetV2 lightweight convolutional neural network model until the success rate of the trained model identifying the test sample set is greater than or equal to the preset success rate.
- test sample set to test the model to verify the model's classification success rate of the sample images in the test sample set. If the classification success rate is greater than or equal to the preset success rate, the training ends, if the classification success rate is less than the preset success rate , After adjusting the relevant training parameters of the model, continue to use the test sample set to train the model until the classification success rate condition is met.
- the embodiment of the application adopts the lightweight convolutional neural network model of the MobileNetV2 network structure.
- the network structure not only has high performance, but also has low complexity.
- the trained model can be directly mounted on mobile electronic devices (such as smart phones). , Recognizing plant images, convenient to carry and use.
- Step 105 Receive the plant image to be recognized, and use the trained model to recognize the type of plant image to be recognized.
- the trained model can be used to recognize the plant image to be recognized, and determine the plant species corresponding to the plant image to be recognized.
- the probability that the image is an Atractylodes lancea image and the probability that the image is an Atractylodes macrocephala image can be obtained.
- FIG. Methods include:
- Step 201 Perform image interception on the plant sample image according to the preset plant parts to obtain the plant sample sub-images corresponding to the preset plant parts.
- the preset plant parts include, but are not limited to, any one or a combination of the following: plant roots, plant leaves, and plant stems.
- the key feature that distinguishes Atractylodes Rhizome and Atractylodes macrocephala is the root feature of the plant, and the root image of the plant can be intercepted on the plant sample image to obtain the root sample sub-image of the plant.
- step 202 if the resolution of the plant sample sub-image is greater than the preset resolution, down-sampling is performed on the plant sample image to obtain a plant sample image of the preset resolution.
- the MobileNetV2 convolutional neural network model used in the embodiments of this application requires an input image with a resolution of 224 ⁇ 224 for training. Therefore, the preset resolution is 224 ⁇ 224, and the resolution of the plant sample sub-image should be 224 ⁇ 224 to ensure The model can train sample images in batches.
- the resolution of the plant sample sub-image is greater than 224 ⁇ 224, the image should be down-sampled to reduce it to 224 ⁇ 224, so that the resolution of all plant sample images can be unified, which is convenient for subsequent operations such as image block processing and establishment Training sample set and test sample set.
- the model includes 53 convolutional layers, a pooling layer, and a fully connected layer that are sequentially connected, wherein the 53 convolutional layers include sequentially connected input layers, 17
- the 53 convolutional layers include sequentially connected input layers, 17
- Each bottleneck building block includes 3 layers of convolutional layers.
- the convolution kernels of the 53 layers of convolutional layers are all 3 ⁇ 3.
- MobileNetV2 is a lightweight convolutional neural network structure.
- the MobileNetV2 model can efficiently identify images with low resolution (for example, images with a resolution of 224 ⁇ 224), and the operation takes up a small bandwidth. It is used on mobile devices.
- Table 1 shows the network structure of the MobileNetV2 model selected in the embodiment of this application. Among them, bottleneck represents the bottleneck building block.
- the model takes an image of 224x224x3 as input, passes through a 1-layer input convolutional layer with a convolution kernel of 3x3 and a step size of 2, and then obtains a 112x112x32 feature map, and then passes through a series of bottleneck modules (convolution of the convolutional layer contained in the module) All the cores are 3x3) to get a 7x7x320 feature map, and then through the output convolution layer of the convolution core 1x1 to get a 7x7x1280 feature map, and then through the 7x7 average pooling layer to get a 1x1x1280 feature map, and finally get a fully connected layer
- the two values are the predicted probabilities of Atractylodes macrocephala and Atractylodes Rhizome.
- those skilled in the art can adjust the number of categories according to the number of plant categories contained in the plant sample image according to actual needs, and adjust the fully connected layer to change its output. This is only an example for illustration.
- Step 203 If the resolution of the plant sample sub-image is less than the preset resolution, perform an upsampling process on the plant sample image to obtain a plant sample image of the preset resolution.
- the resolution of the plant sample sub-image is less than the preset resolution
- the resolution of the image should also be adjusted to the preset resolution.
- the up-sampling process can be used to perform bilinear interpolation up-sampling on the original image To the preset resolution.
- an image with a resolution greater than or equal to a preset resolution is preferred as a sample image to ensure image quality.
- Step 204 According to the RGB value normalization processing formula, the RGB value of each pixel of the plant sample sub-image with the preset resolution is normalized, and the RGB value normalization processing formula is:
- x R , x G , x B are the component values of the R, G, and B channels of any pixel corresponding to the original plant sample sub-image with a preset resolution
- y R , y G , and y B are respectively Are the component values of the R, G, and B channels after normalization of the pixels corresponding to x R , x G , and x B
- x Rmax , x Gmax , and x Bmax are the original plants with the preset resolution respectively The maximum component value on the R, G, and B channels in the sample sub-image.
- Step 205 according to the first preset number of times and the second preset number of times, respectively perform sampling with replacement on the plant sample sub-images of the preset resolution normalized by the RGB values to obtain a training sample set and a test sample set .
- the plant sample sub-image After the plant sample sub-image is normalized by the RGB value, it can be used to establish a training sample set and a test sample set.
- Step 206 Use the training sample set to train the model according to the preset learning rate.
- the model is trained using the training sample set according to the preset learning rate. For example, use the training sample set to train the model at a learning rate of 0.01.
- Step 207 If the success rate of the trained model recognition test sample set is less than the preset success rate, reduce the preset learning rate, and continue training the model according to the reduced preset learning rate until the training model recognizes the test sample set.
- the success rate is greater than or equal to the preset success rate.
- the model After training the model according to the preset learning rate, use the test sample set to test the classification success rate of the model. If the recognition success rate of the model is greater than or equal to the preset success rate, the model has completed training and can be further used to treat The recognized plant images are classified, and if the recognition success rate of the model is less than the preset success rate, the model needs to continue to be trained, and the specific learning rate lower than the original preset learning rate is used for training to improve the recognition of the model Success rate.
- the model is trained 3000 times with a learning rate of 0.01, and each pass contains 30 plant sample images randomly selected from the training sample set. If the trained model recognizes If the success rate is less than 90%, then use a learning rate of 0.001 to train the model 1000 times, each containing 30 plant sample images randomly selected from the training sample set, and then test the recognition success rate of the trained model. If it is still less than 90%, You can reduce the learning rate again to continue learning.
- Step 208 Perform image interception on the received plant image to be recognized according to the preset plant part, to obtain the plant sub-image to be recognized corresponding to the preset plant part;
- Step 209 Perform resolution preprocessing on the plant sub-image to be recognized, so that the resolution of the plant sub-image to be recognized is consistent with the preset resolution;
- Step 210 Perform a normalization process on the RGB value of each pixel of the plant sub-image to be recognized according to the RGB value normalization processing formula
- step 211 the plant sub-images to be recognized after the normalization of the RGB values are input into the trained model to obtain recognition results corresponding to the plant sub-images to be recognized.
- the classification of the plant image to be recognized can be realized.
- the processing procedure is similar to that of the plant sample image.
- the key parts of the image are captured first, and then the captured image is processed into a sub-image with a preset resolution, and then the plant to be recognized
- the RGB value of each pixel of the sub-image is normalized, and finally the processed plant sub-image to be recognized is input into the trained model to obtain the recognition result corresponding to the preset plant part.
- Step 212 If the preset plant part includes one, the recognition result corresponding to the plant sub-image to be recognized is used as the recognition result of the plant image to be recognized.
- the recognition result corresponding to the root image is used as the final recognition result of the plant image to be recognized.
- Step 213 If there are multiple preset plant parts, a weighted sum calculation is performed on multiple corresponding recognition results according to preset weights corresponding to the multiple preset plant parts to obtain the recognition result of the image to be recognized.
- the root, stem, and leaf images of the plant are intercepted, and the recognition results corresponding to the roots, stems, and leaves are obtained, and the recognition results are weighted according to the corresponding preset weights. And calculate the final recognition result of the image to be recognized.
- the difference between the roots is the most obvious relative to the stems and leaves, which should be used as the main reference factor to determine the final recognition result, so its corresponding weight should be higher than the corresponding weights at other positions, thus Get more accurate recognition results.
- the weight corresponding to the root is set to 0.4, and the weight corresponding to the stem and leaves are both 0.3.
- the recognition result corresponding to the root is that the image is atractylodes
- the probability is 80%
- the probability of Atractylodes is 20%
- the probability of atractylodes is 90%
- the identification result corresponding to the leaf is the probability of atractylodes is 90%
- the probability of atractylodes is 30%.
- a rule can be set as follows: if the probability of the recognition result of the first type of plant is greater than or equal to the first preset probability and the probability of the second type of plant is less than the second preset probability, determine the plant corresponding to the plant image to be recognized
- the species is the first kind of plant. For example, the first preset probability is 70% and the second preset probability is 30%, then it can be determined that the type corresponding to the above-mentioned plant image to be recognized is atractylodes. If the recognition result is 71% of Atractylodes and 44% of Atractylodes, then manual judgment can be involved or continue to judge whether the image is another plant image.
- the recognition result is that the probability of atractylodes is 71% and the probability of atractylodes is 44%, and the image is determined to be atractylodes image.
- a plant sample image is intercepted and processed to obtain plant sample sub-images corresponding to preset plant parts, thereby establishing a training sample set and a test sample set, and then using the training sample set to train the classification model.
- the recognition success rate of the test sample set by the classification model meets the preset success rate condition, and a model that can finally be used to classify the plant image to be recognized is obtained.
- the plant image to be recognized is received, the plant image to be recognized is intercepted and processed according to the preset plant parts, and then input into the trained classification model, so as to determine the final recognition result according to the recognition results corresponding to the preset plant parts Classification results of plant images.
- the embodiments of the present application help to solve the problem of low efficiency caused by artificial recognition of plant species in the prior art, and by intercepting the image to retain the characteristic information of the image to remove useless information, the recognition time of the model can be shortened, The identification efficiency of plant species is further improved, and the classification model of the present application can be carried on a mobile device for use, which is convenient for users to carry and use.
- an embodiment of the present application provides a plant image recognition device based on a neural network model.
- the device includes: a sample image interception module 41, sample resolution processing Module 42, sample set establishment module 43, model training module 44, image recognition module 45.
- the sample image interception module 41 is configured to perform image interception on the plant sample image according to the preset plant parts to obtain the plant sample sub-images corresponding to the preset plant parts;
- the sample resolution processing module 42 is configured to perform resolution preprocessing on the plant sample sub-image to obtain a plant sample image of the first preset resolution
- the sample set establishment module 43 is configured to perform sampling with replacement on the plant sample sub-images of the preset resolution according to the first preset number of times and the second preset number of times to obtain a training sample set and a test sample set.
- the first preset number of times is greater than the second preset number of times;
- the model training module 44 is used to train the MobileNetV2 lightweight convolutional neural network model by using the training sample set until the success rate of the trained model recognition test sample set is greater than or equal to the preset success rate;
- the image recognition module 45 is used to receive the plant image to be recognized, and use the trained model to recognize the type of the plant image to be recognized.
- the first preset resolution is greater than the preset resolution;
- the sample resolution processing module 42 specifically includes: a sample down-sampling unit 421 and a sample up-sampling unit 422.
- the sample down-sampling unit 421 is configured to perform down-sampling processing on the plant sample image if the resolution of the plant sample sub-image is greater than the preset resolution to obtain a plant sample image of the preset resolution;
- the sample up-sampling unit 422 is configured to perform an up-sampling process on the plant sample image if the resolution of the plant sample sub-image is less than the preset resolution to obtain a plant sample image of the preset resolution.
- the device further includes: an RGB value normalization module 46.
- the RGB value normalization module 46 is used to preprocess the resolution of the plant sample sub-images, and after obtaining the plant sample sub-images of the preset resolution, according to the RGB value normalization processing formula, the plant samples of the preset resolution
- the RGB value of each pixel of the sub-image is normalized, and the RGB value normalization processing formula is:
- x R , x G , x B are the component values of the R, G, and B channels of any pixel corresponding to the original plant sample sub-image with a preset resolution
- y R , y G , and y B are respectively Are the component values of the R, G, and B channels after normalization of the pixels corresponding to x R , x G , and x B
- x Rmax , x Gmax , and x Bmax are the original plants with the preset resolution respectively The maximum component value on the R, G, and B channels in the sample sub-image.
- the sample set establishment module 43 is specifically configured to establish a training sample set and a test sample set by using the plant sample sub-images of the preset resolution after normalization of the RGB values.
- the image recognition module 45 specifically includes: an image interception unit 451, an image resolution processing unit 452, an image RGB value normalization unit 453, an image recognition unit 454, and recognition result calculation Unit 455.
- the image resolution processing unit 451 is configured to perform image interception according to preset plant parts on the received plant image to be recognized, to obtain the plant sub-image to be recognized corresponding to the preset plant part;
- the image segmentation unit 452 is configured to perform resolution preprocessing on the plant sub-image to be recognized, so that the resolution of the plant sub-image to be recognized is consistent with the preset resolution;
- the image RGB value normalization unit 453 is configured to perform normalization processing on the RGB value of each pixel of the plant sub-image to be recognized according to the RGB value normalization processing formula;
- the image recognition unit 454 is configured to input the plant sub-image to be recognized after the normalization of RGB values into the trained model to obtain a recognition result corresponding to the plant sub-image to be recognized;
- the recognition result calculation unit 455 is configured to, if the preset plant part includes one, use the recognition result corresponding to the plant sub-image to be recognized as the recognition result of the plant image to be recognized;
- a weighted sum calculation is performed on the multiple corresponding recognition results according to the preset weights corresponding to the multiple preset plant parts to obtain the recognition result of the image to be recognized.
- the model training module 44 specifically includes: a first model training unit 441 and a second model training unit 442.
- the first model training unit 441 is configured to use the training sample set to train the model according to the preset learning rate
- the second model training unit 442 is configured to reduce the preset learning rate if the success rate of the trained model recognition test sample set is less than the preset success rate, and continue training the model according to the reduced preset learning rate until the training The success rate of the model identification test sample set is greater than or equal to the preset success rate.
- the network structure of the model includes sequentially connected 53 layers of convolutional layers, 1 layer of pooling layers, and 1 layer of fully connected layers, where the 53 layers of convolutional layers include sequentially connected 1 layer of input layers , 17 bottleneck building blocks, 1 layer output layer, each bottleneck building block includes 3 layers of convolutional layers, and the convolution kernels of the 53 layers of convolutional layers are all 3 ⁇ 3.
- an embodiment of the present application also provides a computer-readable storage medium on which computer-readable instructions are stored.
- the program When the program is executed by a processor, the following steps are implemented: Performing image interception on the image according to preset plant parts to obtain plant sample sub-images corresponding to said preset plant parts; performing resolution preprocessing on said plant sample sub-images to obtain plant sample sub-images with preset resolution; According to the first preset number of times and the second preset number of times, the plant sample sub-images of the preset resolution are respectively sampled with replacement to obtain a training sample set and a test sample set, wherein the first preset The number of times is greater than the second preset number of times; use the training sample set to train the MobileNetV2 lightweight convolutional neural network model until the trained model recognizes that the success rate of the test sample set is greater than or equal to the preset success rate; receive The plant image to be recognized, and the type of the plant image to be recognized is recognized by using the trained model.
- the embodiment of the present application also provides a physical structure diagram of a computer device, as shown in FIG. 5,
- the computer equipment includes: a processor 41, a memory 42, and computer readable instructions stored on the memory 42 and running on the processor, wherein the memory 42 and the processor 41 are both set on the bus 43 and the processor 41 executes
- the program implements the following steps: performing image interception on the plant sample image according to preset plant parts to obtain plant sample sub-images corresponding to the preset plant parts; performing resolution preprocessing on the plant sample sub-images, Obtain the plant sample sub-images of the preset resolution; according to the first preset number of times and the second preset number of times, perform sampling with replacement on the plant sample sub-images of the preset resolution to obtain a training sample set and a test A sample set, wherein the first preset number of times is greater than the second preset number of times; the MobileNetV2 lightweight con
- the technical solution of this application can be embodied in the form of a software product.
- the software product can be stored in a non-volatile storage medium (which can be a CD-ROM, U disk, mobile hard disk, etc.), including several
- the instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute the methods described in each implementation scenario of this application.
- a computer device which may be a personal computer, a server, or a network device, etc.
- the resolution preprocessing is performed on the plant sample sub-images obtained by the interception, and the plant sample sub-images after the resolution preprocessing are used to establish a training sample set and a test sample set, so as to use the training sample set to compare the corresponding MobileNetV2 convolutional neural
- the network model is trained so that the recognition success rate of the trained model for the test sample set reaches the preset success rate standard, so that the new plant image to be recognized can be classified and the plant species can be recognized through the trained model.
- this application can use images of plants to be identified taken by mobile devices equipped with trained models to identify the types of plants through the model, which improves the The efficiency of type recognition saves labor costs.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
本申请公开了一种基于神经网络模型的植物图像的识别方法及装置、存储介质、计算机设备,该方法包括:在植物样本图像上按照预设植物部位进行图像截取,得到与预设植物部位对应的植物样本子图像;对植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本子图像;按照第一预设次数和第二预设次数,分别对所述预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本;利用训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别测试样本集的成功率大于或等于预设成功率;接收待识别的植物图像,并利用训练好的模型对待识别的植物图像的种类进行识别。本申请有助于提高植物种类的识别效率,节约人力成本。
Description
本申请要求与2019年3月12日提交中国专利局、申请号为2019101850769、申请名称为“植物种类的识别方法及装置、存储介质、计算机设备”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
本申请涉及图像识别技术领域,尤其是涉及到一种基于神经网络模型的植物图像的识别方法及装置、存储介质、计算机设备。
如何识别植物种类是广大种植者的需求,在种植过程中,种植者需要去除其他种类植株,只在种植区域内保留所需种类植株。比如在苍术的种植过程中,经常会混入与苍术同属菊科苍术属的白术,两种植物的药效不同,但是外形十分相似,在苍术的种植过程中,需要去除白术和其他植株,防止白术或其他植株混入苍术中影响苍术药效。
对于外形相似的植株,目前需要经验丰富的人仔细观察进行区分,但是人工分类速度较慢并且对人员经验要求较高,人工分类的方式不仅所需人力成本较高,而且很难同时兼顾产量和品质需求。
发明内容
有鉴于此,本申请提供了一种基于神经网络模型的植物图像的识别方法及装置、存储介质、计算机设备,有助于提高植物种类的识别效率。
根据本申请的一个方面,提供了一种基于神经网络模型的植物图像的识别方法,包括:
在植物样本图像上按照预设植物部位进行图像截取,得到与所述预设植物部位对应的植物样本子图像;
对所述植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本子图像;
按照第一预设次数和第二预设次数,分别对所述预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本集,其中,所述第一预设次数大于所述第二预设次数;
利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别所述测试样本集的成功率大于或等于预设成功率;
接收待识别的植物图像,并利用所述训练好的模型对所述待识别的植物图像的种类进 行识别。
根据本申请的另一方面,提供了一种基于神经网络模型的植物图像的识别装置,包括:
样本图像截取模块,用于在植物样本图像上按照预设植物部位进行图像截取,得到与所述预设植物部位对应的植物样本子图像;
样本分辨率处理模块,用于对所述植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本图像;
样本集建立模块,用于按照第一预设次数和第二预设次数,分别对所述预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本集,其中,所述第一预设次数大于所述第二预设次数;
模型训练模块,用于利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别所述测试样本集的成功率大于或等于预设成功率;
图像识别模块,用于接收待识别的植物图像,并利用所述训练好的模型对所述待识别的植物图像的种类进行识别。
依据本申请又一个方面,提供了一种存储介质,其上存储有计算机程序,所述程序被处理器执行时实现上述基于神经网络模型的植物图像的识别方法。
依据本申请再一个方面,提供了一种计算机设备,包括存储介质、处理器及存储在存储介质上并可在处理器上运行的计算机程序,所述处理器执行所述程序时实现上述基于神经网络模型的植物图像的识别方法。
借由上述技术方案,本申请提供的一种基于神经网络模型的植物图像的识别方法及装置、存储介质、计算机设备,截取与预设植物部位对应的植物样本子图像后,对截取得到的植物样本子图像进行分辨率预处理,并利用分辨率预处理后的植物样本子图像建立训练样本集以及测试样本集,从而利用训练样本集对相应的MobileNetV2卷积神经网络模型进行训练,使得训练后的模型对测试样本集的识别成功率达到预设成功率的标准,从而可以通过训练后的模型对新的待识别的植物图像的分类,识别植物种类。本申请与现有技术中依靠有经验的技术人员对植物进行人工分类相比,可以利用搭载训练后的模型的移动设备拍摄的待识别植物图像,通过模型对植物的种类进行识别,提高了植物种类的识别效率,节约了人力成本。
上述说明仅是本申请技术方案的概述,为了能够更清楚了解本申请的技术手段,而可依照说明书的内容予以实施,并且为了让本申请的上述和其它目的、特征和优点能够更明显易懂,以下特举本申请的具体实施方式。
此处所说明的附图用来提供对本申请的进一步理解,构成本申请的一部分,本申请的示意性实施例及其说明用于解释本申请,并不构成对本申请的不当限定。在附图中:
图1示出了本申请实施例提供的一种基于神经网络模型的植物图像的识别方法的流程示意图;
图2示出了本申请实施例提供的另一种基于神经网络模型的植物图像的识别方法的流程示意图;
图3示出了本申请实施例提供的一种基于神经网络模型的植物图像的识别装置的结构示意图;
图4示出了本申请实施例提供的另一种基于神经网络模型的植物图像的识别装置的结构示意图;
图5示出了本申请实施例提供的一种计算机设备的实体结构示意图。
下文中将参考附图并结合实施例来详细说明本申请。需要说明的是,在不冲突的情况下,本申请中的实施例及实施例中的特征可以相互组合。
在本实施例中提供了一种基于神经网络模型的植物图像的识别方法,如图1所示,该方法包括:
步骤101,在植物样本图像上按照预设植物部位进行图像截取,得到与预设植物部位对应的植物样本子图像。
本申请的实施例以区分苍术图像和白术图像为例进行解释,本申请的实施例仅作为举例说明,本领域技术人员可以将苍术图像、白术图像替换为其他的一种或多种植物图像,在此不做限定。
在植物的识别过程中,通常是根据植物的某个部位的特征进行识别,苍术与白术的叶子比较相似,而根部特征有所不同,则对于苍术、白术样本图像可以截取植株的根部图像,以便利用根部特征进行训练。通常截取的图像包括原有的植物样本图像的一部分,选取植物的关键特征忽略次要特征,从而利用截取后的图像进行训练,避免损失图像的关键特征信息的同时可以加快训练速度,提升建模效率。
步骤102,对植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本图像。
在上述实施例中,对植物样本图像进行截取得到植物样本子图像后,为了方便利用样本子图像训练分类模型,提高训练效率,需要对样本子图像的分辨率进行归一化,将样本子图像的大小统一化,具体可以将全部的植物样本子图像处理为预设分辨率的图像。
步骤103,按照第一预设次数和第二预设次数,分别对预设分辨率的植物样本子图像进行有放回的随机抽样,得到训练样本集和测试样本集,其中,第一预设次数大于第二预设次数。
本申请实施例采用Bootstrap方法建立训练样本集和测试样本集,分别按照第一预设次数和第二预设次数对分辨率处理后的植物样本子图像进行有放回的抽样,从而建立训练样 本集和测试样本集,其中,训练样本集和测试样本集中都包含有多张带有苍术标签的苍术样本子图像和带有白术标签的白术样本子图像,并且一般来说训练样本集的样本数量要高于测试样本集的样本数量,即第一预设次数大于第二预设次数,例如本实施例中可以对植物样本子图像进行2000次有放回的抽样建立训练样本集,进行300次有放回的抽样建立测试样本集。
需要说明的是,抽样可以采用随机抽样的方式,也可以采用分层抽样的方式。
步骤104,利用训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别测试样本集的成功率大于或等于预设成功率。
利用测试样本集对模型进行测试,以验证模型对测试样本集中的样本图像的分类成功率,如果分类成功率大于或等于预设的成功率,则训练结束,如果分类成功率小于预设成功率,则调整模型的相关训练参数后,继续利用测试样本集对模型进行训练,直至满足分类成功率条件为止。
本申请实施例采用MobileNetV2网络结构的轻量级卷积神经网络模型,该网络结构不仅具有高性能,而且复杂度较低,训练后的模型可以直接搭载在移动电子设备(如智能手机)上使用,进行植物图像的识别,方便携带和使用。
步骤105,接收待识别的植物图像,并利用训练好的模型对待识别的植物图像的种类进行识别。
模型训练成功后,就可以通过训练好的模型对待识别的植物图像进行识别,确定待识别的植物图像对应的植物种类。本申请的实施例中,向模型输入待识别的植物图像后,可以得出该图像为苍术图像的概率和该图像为白术图像的概率。
通过应用本实施例的技术方案,截取与预设植物部位对应的植物样本子图像后,对截取得到的植物样本子图像进行分辨率预处理,并利用分辨率预处理后的植物样本子图像建立训练样本集以及测试样本集,从而利用训练样本集对相应的MobileNetV2卷积神经网络模型进行训练,使得训练后的模型对测试样本集的识别成功率达到预设成功率的标准,从而可以通过训练后的模型对新的待识别的植物图像的分类,识别植物种类。本申请与现有技术中依靠有经验的技术人员对植物进行人工分类相比,可以利用搭载训练后的模型的移动设备拍摄的待识别植物图像,通过模型对植物的种类进行识别,提高了植物种类的识别效率,节约了人力成本。
进一步的,作为上述实施例具体实施方式的细化和扩展,为了完整说明本实施例的具体实施过程,提供了另一种基于神经网络模型的植物图像的识别方法,如图2所示,该方法包括:
步骤201,在植物样本图像上按照预设植物部位进行图像截取,得到与预设植物部位对应的植物样本子图像。
在上述实施例中,具体地,预设植物部位包括但不限于以下任意一种或其组合:植物根部、植物叶子、植物茎部。
例如区分苍术和白术植株的关键特征为植株根本特征,则可以在植物样本图像上对植株的根部图像进行截取,得倒植物的根部样本子图像。
步骤202,若植物样本子图像的分辨率大于预设分辨率,则对植物样本图像进行降采样处理,得到预设分辨率的植物样本图像。
本申请实施例采用的MobileNetV2卷积神经网络模型需要输入分辨率为224×224的图像进行训练,因此预设分辨率取224×224,植物样本子图像的分辨率应为224×224,以保证模型能够对样本图像进行批量训练。
若植物样本子图像的分辨率大于224×224,应对图像进行降采样处理使其下降到224×224,使全部的植物样本图像的分辨率得到统一,方便后续的图像分块处理等操作以及建立训练样本集和测试样本集。
在本申请实施例中,具体地,模型包括依次连接的53层卷积层、1层池化层以及1层全连接层,其中,53层卷积层包括依次连接的1层输入层、17个瓶颈构建块、1层输出层,每个瓶颈构建块分别包括3层卷积层,53层卷积层的卷积核均为3×3。
MobileNetV2是一种轻量级的卷积神经网络结构,MobileNetV2模型可以高效的对分辨率不高的图像(例如分辨率为224×224的图像)进行快速的识别,并且运算占用带宽较小,可以搭载于移动设备上使用。表1示出了本申请实施例选用的MobileNetV2模型的网络结构。其中,bottleneck表示瓶颈构建块。模型以224x224x3的图像作为输入,经过卷积核为3x3、步长为2的1层输入卷积层后,得到112x112x32的特征图,然后经过一系列bottleneck模块(模块中包含的卷积层的卷积核全部为3x3)得到7x7x320的特征图,之后通过卷积核为1x1的输出卷积层得到7x7x1280的特征图,再通过7x7的平均池化层得到1x1x1280的特征图,最后通过全连接层得到2个数值,分别为白术和苍术各自的概率预测值。当然,本领域技术人员可以根据实际需要根据植物样本图像中包含的植物类别的数量调整类别数量,对全连接层进行调整改变其输出,在此仅作为举例说明。
表1 MobileNetV2网络结构组成表
步骤203,若植物样本子图像的分辨率小于预设分辨率,则对植物样本图像进行升采样处理,得到预设分辨率的植物样本图像。
如果植物样本子图像的分辨率小于预设分辨率,也应将该图像的分辨率调整为预设分辨率大小,具体可以采用升采样处理方式,对原有的图像进行双线性插值上采样至预设分辨率。
需要说明的是,对图像进行升采样处理可能会对图像的性能造成一定程度的损失,因此,优选分辨率大于或等于的预设分辨率的图像作为样本图像,保证图像的质量。
步骤204,按照RGB值归一化处理公式,将预设分别率的植物样本子图像的每个像素点的RGB值进行归一化处理,RGB值归一化处理公式为:
其中,x
R、x
G、x
B分别为原有的预设分辨率的植物样本子图像对应的任一像素点的R、G、B通道的分量值,y
R、y
G、y
B分别为与x
R、x
G、x
B对应的像素点经过归一化处理后的R、G、B通道的分量值,x
Rmax、x
Gmax、x
Bmax分别为原有的预设分辨率的植物样本子图像中在R、G、B通道上的最大分量值。
在建立训练样本集和测试样本集之前,还需要利用上述RGB值归一化处理公式对图像截取得到的植物样本子图像进行RGB值归一化处理,使样本子图像的每个像素点在R、G、B三通道上的分量值变换到[0,1]之间,使数据分布更加均匀,从而有助于加速模型的训练过程。
例如,原有全部样本子图像中的R通道最大分量值为200,对其中一张子图像进行RGB值归一化处理,待处理的子图像的R通道分量值为40,那么归一化处理后的该子图像的R通道分量值应为40/200=0.2,然后再按照上述公式对G和B通道的分量值进行归一化处理。
步骤205,按照第一预设次数和第二预设次数,分别对RGB值归一化处理后的预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本集。
植物样本子图像经过RGB值归一化处理后,则可以用于建立训练样本集和测试样本集。
步骤206,利用训练样本集,按照预设学习率对模型进行训练。
按照预先设定的学习率对模型利用训练样本集对模型进行训练。例如按照0.01的学习率利用训练样本集训练模型。
步骤207,若训练后的模型识别测试样本集的成功率小于预设成功率,则降低预设学习率,并按照降低后的预设学习率继续训练模型直至训练后的模型识别测试样本集的成功率大于或等于预设成功率为止。
按照预设学习率对模型进行训练后,利用测试样本集测试该模型的分类成功率,如果模型的识别成功率大于或等于预设的成功率,则该模型完成训练,可以进一步利用该模型对待识别的植物图像进行分类,而如果模型的识别成功率小于预设成功率,则需要继续对模型进行训练,具体应用比原来的预设学习率更低的学习率进行训练,从而提高模型的识别成功率。
例如,假设预设学习率为0.01,预设成功率为90%,先使用0.01的学习率训练模型3000遍,每遍包含训练样本集中随机抽取的30张植物样本图像,如果训练后的模型识别成功率小于90%,则再使用0.001的学习率训练模型1000遍,每遍包含训练样本集中随机抽取的30张植物样本图像,再测试训练后的模型的识别成功率,如果仍然小于90%,可以再次降低学习率继续学习。
步骤208,在接收到的待识别的植物图像上按照预设植物部位进行图像截取,得到与预设植物部位对应的待识别的植物子图像;
步骤209,对待识别的植物子图像进行分辨率预处理,以使待识别的植物子图像的分辨率与预设分辨率一致;
步骤210,按照RGB值归一化处理公式,对待识别的植物子图像的每个像素点的RGB值进行归一化处理;
步骤211,分别将RGB值归一化处理后的待识别的植物子图像输入至训练好的模型中,得到与待识别的植物子图像对应的识别结果。
在上述步骤208至步骤211中,模型训练完成后,可以实现对待识别的植物图像的分类。具体地,接收到待识别的植物图像后,与植物样本图像的处理过程相似,先对图像进行关键部位截取,然后将截取后的图像处理为预设分辨率的子图像,再对待识别的植物子图像的每个像素点的RGB值进行归一化处理,最后将处理后的待识别的植物子图像输入至训练好的模型中得到与预设植物部位对应的识别结果。
步骤212,若预设植物部位包括一个,则将与待识别的植物子图像对应的识别结果作为待识别的植物图像的识别结果。
如果预设植物部位包括一个,例如只截取了根部图像,则将根部图像对应的识别结果作为待识别的植物图像的最终识别结果。
步骤213,若预设植物部位包括多个,则按照与多个预设植物部位对应的预设权重,对多个相应的识别结果进行加权求和计算,得到待识别的图像的识别结果。
如果预设植物部位包括多个,例如截取了植物的根部、茎部和叶子图像,则分别得到与根部、茎部和叶子对应的识别结果后,按照相应的预设权重将识别结果进行加权求和计算得到待识别的图像的最终识别结果。其中,对于苍术和白术的识别来说,根部的区别相对于茎部和叶子最为明显,应作为确定最终识别结果的主要参考因素,因此其对应的权重应高于其他位置处对应的权重,从而得到更准确的识别结果。
例如,根部对应的权重设为0.4,茎部和叶子对应的权重都为0.3,假设根部对应的识别结果为图像为苍术的概率是80%,白术概率是20%,茎部对于的识别结果为苍术概率为40%,白术概率为90%,叶子对应的识别结果为苍术概率90%,白术概率30%。则最终得到的待识别的植物图像为苍术的概率为80%*0.4+40%*0.3+90%*0.3=71%,待识别的植物图像为白术的概率为20%*0.4+90%*0.3+30%*0.3=44%。
具体地,可以设置规定为:若识别结果为第一种类植物的概率大于或等于第一预设概率且第二种类植物的概率小于第二预设概率,则确定待识别的植物图像对应的植物种类为第一种类植物。例如第一预设概率为70%,第二预设概率为30%,那么可以判定上述的待识别的植物图像对应的种类为苍术。而如果识别结果为苍术概率71%,白术概率44%,则可以介入人工判断或继续判断该图像是否为其他植物图像。
另外,也可以比较两种植物识别结果的植物种类对应的概率大小,将概率大的植物种类确定为待识别的植物图像的种类。例如识别结果为苍术概率71%,白术概率44%,则判定该图像为苍术图像。
通过应用本实施例的技术方案,将一张植物样本图像进行截取处理得到与预设植物部位对应的植物样本子图像,从而建立训练样本集和测试样本集,进而分别利用训练样本集训练分类模型,使得分类模型对测试样本集的识别成功率满足预设的成功率条件,得到最终可以用于对待识别的植物图像进行分类的模型。当接收到待识别的植物图像时,将待识别的植物图像按照预设植物部位截取处理后,输入至训练好的分类模型中,从而根据与预设植物部位对应的识别结果确定最终的待识别植物图像的分类结果。本申请的实施例有助于解决现有技术中依靠人工进行植物种类识别导致的效率低下的问题,并且通过对图像进行截取处理以保留图像的特征信息去除无用信息,能够缩短模型的识别时间,进一步的提高了植物种类的识别效率,同时本申请的分类模型可以搭载在移动设备上使用,方便用户随身携带和使用。
进一步的,作为图1方法的具体实现,本申请实施例提供了一种基于神经网络模型的植物图像的识别装置,如图3所示,该装置包括:样本图像截取模块41,样本分辨率处理模块42、样本集建立模块43、模型训练模块44、图像识别模块45。
样本图像截取模块41,用于在植物样本图像上按照预设植物部位进行图像截取,得到与预设植物部位对应的植物样本子图像;
样本分辨率处理模块42,用于对植物样本子图像进行分辨率预处理,得到第一预设分辨率的植物样本图像;
样本集建立模块43,用于按照第一预设次数和第二预设次数,分别对预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本集,其中,第一预设次数大于第二预设次数;
模型训练模块44,用于利用训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别测试样本集的成功率大于或等于预设成功率;
图像识别模块45,用于接收待识别的植物图像,并利用训练好的模型对待识别的植物图像的种类进行识别。
在具体的应用场景中,如图4所示,第一预设分辨率大于预设分辨率;样本分辨率处理模块42,具体包括:样本降采样单元421、样本升采样单元422。
样本降采样单元421,用于若植物样本子图像的分辨率大于预设分辨率,则对植物样本图像进行降采样处理,得到预设分辨率的植物样本图像;
样本升采样单元422,用于若植物样本子图像的分辨率小于预设分辨率,则对植物样本图像进行升采样处理,得到预设分辨率的植物样本图像。
在具体的应用场景中,如图4所示,该装置还包括:RGB值归一化模块46。
RGB值归一化模块46,用于对植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本子图像之后,按照RGB值归一化处理公式,将预设分别率的植物样本子图像的每个像素点的RGB值进行归一化处理,RGB值归一化处理公式为:
其中,x
R、x
G、x
B分别为原有的预设分辨率的植物样本子图像对应的任一像素点的R、G、B通道的分量值,y
R、y
G、y
B分别为与x
R、x
G、x
B对应的像素点经过归一化处理后的R、G、B通道的分量值,x
Rmax、x
Gmax、x
Bmax分别为原有的预设分辨率的植物样本子图像中在R、G、B通道上的最大分量值。
样本集建立模块43,具体用于利用RGB值归一化处理后的预设分辨率的植物样本子图像,建立训练样本集和测试样本集。
在具体的应用场景中,如图4所示,图像识别模块45,具体包括:图像截取单元451,图像分辨率处理单元452,图像RGB值归一化单元453,图像识别单元454,识别结果计算单元455。
图像分辨率处理单元451,用于在接收到的待识别的植物图像上按照预设植物部位进行图像截取,得到与预设植物部位对应的待识别的植物子图像;
图像分块单元452,用于对待识别的植物子图像进行分辨率预处理,以使待识别的植物子图像的分辨率与预设分辨率一致;
图像RGB值归一化单元453,用于按照RGB值归一化处理公式,分别对待识别的植物子图像的每个像素点的RGB值进行归一化处理;
图像识别单元454,用于将RGB值归一化处理后的待识别的植物子图像输入至训练好的模型中,得到与待识别的植物子图像对应的识别结果;
识别结果计算单元455,用于若预设植物部位包括一个,则将与待识别的植物子图像对应的识别结果作为待识别的植物图像的识别结果;
若预设植物部位包括多个,则按照与多个预设植物部位对应的预设权重,对多个相应的识别结果进行加权求和计算,得到待识别的图像的识别结果。
在具体的应用场景中,如图4所示,模型训练模块44,具体包括:第一模型训练单元441、第二模型训练单元442。
第一模型训练单元441,用于利用训练样本集,按照预设学习率对模型进行训练;
第二模型训练单元442,用于若训练后的模型识别测试样本集的成功率小于预设成功率,则降低预设学习率,并按照降低后的预设学习率继续训练模型直至训练后的模型识别测试样本集的成功率大于或等于预设成功率为止。
在上述实施例中,具体地,模型的网络结构包括依次连接的53层卷积层、1层池化层以及1层全连接层,其中,53层卷积层包括依次连接的1层输入层、17个瓶颈构建块、1层输出层,每个瓶颈构建块分别包括3层卷积层,53层卷积层的卷积核均为3×3。
需要说明的是,本申请实施例提供的一种基于神经网络模型的植物图像的识别装置所涉及各功能单元的其他相应描述,可以参考图1和图2中的对应描述,在此不再赘述。
基于上述如图1所示方法,相应的,本申请实施例还提供了一种计算机可读存储介质,其上存储有计算机可读指令,该程序被处理器执行时实现以下步骤:在植物样本图像上按照预设植物部位进行图像截取,得到与所述预设植物部位对应的植物样本子图像;对所述植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本子图像;按照第一预设次数和第二预设次数,分别对所述预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本集,其中,所述第一预设次数大于所述第二预设次数;利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别所述测试样本集的成功率大于或等于预设成功率;接收待识别的植物图像,并利用所述训练好的模型对所述待识别的植物图像的种类进行识别。
基于上述如图1所示方法和如图3所示基于神经网络模型的植物图像的识别装置的实施例,本申请实施例还提供了一种计算机设备的实体结构图,如图5所示,该计算机设备包括:处理器41、存储器42、及存储在存储器42上并可在处理器上运行的计算机可读指令,其中存储器42和处理器41均设置在总线43上所述处理器41执行所述程序时实现以下步骤:在植物样本图像上按照预设植物部位进行图像截取,得到与所述预设植物部位对应的植物样本子图像;对所述植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本子图像;按照第一预设次数和第二预设次数,分别对所述预设分辨率的植物样本子图 像进行有放回的抽样,得到训练样本集和测试样本集,其中,所述第一预设次数大于所述第二预设次数;利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别所述测试样本集的成功率大于或等于预设成功率;接收待识别的植物图像,并利用所述训练好的模型对所述待识别的植物图像的种类进行识别。
基于这样的理解,本申请的技术方案可以以软件产品的形式体现出来,该软件产品可以存储在一个非易失性存储介质(可以是CD-ROM,U盘,移动硬盘等)中,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行本申请各个实施场景所述的方法。通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到本申请可以借助软件加必要的通用硬件平台的方式来实现,也可以通过硬件实现截取与预设植物部位对应的植物样本子图像后,对截取得到的植物样本子图像进行分辨率预处理,并利用分辨率预处理后的植物样本子图像建立训练样本集以及测试样本集,从而利用训练样本集对相应的MobileNetV2卷积神经网络模型进行训练,使得训练后的模型对测试样本集的识别成功率达到预设成功率的标准,从而可以通过训练后的模型对新的待识别的植物图像的分类,识别植物种类。本申请与现有技术中依靠有经验的技术人员对植物进行人工分类相比,可以利用搭载训练后的模型的移动设备拍摄的待识别植物图像,通过模型对植物的种类进行识别,提高了植物种类的识别效率,节约了人力成本。
本领域技术人员可以理解附图只是一个优选实施场景的示意图,附图中的模块或流程并不一定是实施本申请所必须的。本领域技术人员可以理解实施场景中的装置中的模块可以按照实施场景描述进行分布于实施场景的装置中,也可以进行相应变化位于不同于本实施场景的一个或多个装置中。上述实施场景的模块可以合并为一个模块,也可以进一步拆分成多个子模块。
上述本申请序号仅仅为了描述,不代表实施场景的优劣。以上公开的仅为本申请的几个具体实施场景,但是,本申请并非局限于此,任何本领域的技术人员能思之的变化都应落入本申请的保护范围。
Claims (20)
- 一种基于神经网络模型的植物图像的识别方法,其特征在于,包括:在植物样本图像上按照预设植物部位进行图像截取,得到与所述预设植物部位对应的植物样本子图像;对所述植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本子图像;按照第一预设次数和第二预设次数,分别对所述预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本集,其中,所述第一预设次数大于所述第二预设次数;利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别所述测试样本集的成功率大于或等于预设成功率;接收待识别的植物图像,并利用所述训练好的模型对所述待识别的植物图像的种类进行识别。
- 根据权利要求1所述的方法,其特征在于,所述利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别所述测试样本集的成功率大于或等于预设成功率,具体包括:利用所述训练样本集,按照预设学习率对所述模型进行训练;若训练后的所述模型识别所述测试样本集的成功率小于所述预设成功率,则降低所述预设学习率,并按照降低后的预设学习率继续训练所述模型直至训练后的所述模型识别所述测试样本集的成功率大于或等于所述预设成功率为止。
- 根据权利要求1所述的方法,其特征在于,所述模型的网络结构包括依次连接的53层卷积层、1层池化层以及1层全连接层,其中,53层卷积层包括依次连接的1层输入层、17个瓶颈构建块、1层输出层,每个瓶颈构建块分别包括3层卷积层,53层卷积层的卷积核均为3×3。
- 根据权利要求1至3中任一项所述的方法,其特征在于,所述对所述植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本子图像,具体包括:若所述植物样本子图像的分辨率大于所述预设分辨率,则对所述植物样本图像进行降采样处理,得到所述预设分辨率的植物样本图像;若所述植物样本子图像的分辨率小于所述预设分辨率,则对所述植物样本图像进行升采样处理,得到所述预设分辨率的植物样本图像。
- 根据权利要求4所述的方法,其特征在于,所述对所述植物样本子图像进行分辨 率预处理,得到预设分辨率的植物样本子图像之后,所述方法还包括:按照RGB值归一化处理公式,将所述预设分别率的植物样本子图像的每个像素点的RGB值进行归一化处理,所述RGB值归一化处理公式包括:其中,x R、x G、x B分别为原有的所述预设分辨率的植物样本子图像对应的任一像素点的R、G、B通道的分量值,y R、y G、y B分别为与x R、x G、x B对应的像素点经过归一化处理后的R、G、B通道的分量值,x Rmax、x Gmax、x Bmax分别为原有的所述预设分辨率的植物样本子图像中在R、G、B通道上的最大分量值;所述按照第一预设次数和第二预设次数,分别对所述预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本集,具体包括:按照第一预设次数和第二预设次数,分别对RGB值归一化处理后的所述预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本集。
- 根据权利要求5所述的方法,其特征在于,所述预设植物部位包括但不限于以下任意一种或其组合:植物根部、植物叶子、植物茎部。
- 根据权利要求6所述的方法,其特征在于,所述接收待识别的植物图像,并利用所述训练好的模型对所述待识别的植物图像的种类进行识别,具体包括:在接收到的待识别的植物图像上按照所述预设植物部位进行图像截取,得到与所述预设植物部位对应的待识别的植物子图像;对所述待识别的植物子图像进行分辨率预处理,以使所述待识别的植物子图像的分辨率与所述预设分辨率一致;按照所述RGB值归一化处理公式,对所述待识别的植物子图像的每个像素点的RGB值进行归一化处理;将RGB值归一化处理后的所述待识别的植物子图像输入至所述训练好的模型中,得到与所述待识别的植物子图像对应的识别结果;若所述预设植物部位包括一个,则将与所述待识别的植物子图像对应的识别结果作为所述待识别的植物图像的识别结果;若所述预设植物部位包括多个,则按照与多个所述预设植物部位对应的预设权重,对多个相应的所述识别结果进行加权求和计算,得到所述待识别的图像的识别结果。
- 一种基于神经网络模型的植物图像的识别装置,其特征在于,包括:样本图像截取模块,用于在植物样本图像上按照预设植物部位进行图像截取,得到与 所述预设植物部位对应的植物样本子图像;样本分辨率处理模块,用于对所述植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本图像;样本集建立模块,用于按照第一预设次数和第二预设次数,分别对所述预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本集,其中,所述第一预设次数大于所述第二预设次数;模型训练模块,用于利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别所述测试样本集的成功率大于或等于预设成功率;图像识别模块,用于接收待识别的植物图像,并利用所述训练好的模型对所述待识别的植物图像的种类进行识别。
- 根据权利要求8所述的装置,其特征在于,所述模型训练模块,具体包括:第一模型训练单元,用于利用所述训练样本集,按照预设学习率对所述模型进行训练;第二模型训练单元,用于若训练后的所述模型识别所述测试样本集的成功率小于所述预设成功率,则降低所述预设学习率,并按照降低后的预设学习率继续训练所述模型直至训练后的所述模型识别所述测试样本集的成功率大于或等于所述预设成功率为止。
- 根据权利要求8所述的装置,其特征在于,所述模型的网络结构包括依次连接的53层卷积层、1层池化层以及1层全连接层,其中,53层卷积层包括依次连接的1层输入层、17个瓶颈构建块、1层输出层,每个瓶颈构建块分别包括3层卷积层,53层卷积层的卷积核均为3×3。
- 根据权利要求8至10中任一项所述的装置,其特征在于,所述样本分辨率处理模块,具体包括:样本降采样单元,用于若所述植物样本子图像的分辨率大于所述预设分辨率,则对所述植物样本图像进行降采样处理,得到所述预设分辨率的植物样本图像;样本升采样单元,用于若所述植物样本子图像的分辨率小于所述预设分辨率,则对所述植物样本图像进行升采样处理,得到所述预设分辨率的植物样本图像。
- 根据权利要求11所述的装置,其特征在于,所述装置还包括:RGB值归一化模块,用于对所述植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本子图像之后,按照RGB值归一化处理公式,将所述预设分别率的植物样本子图像的每个像素点的RGB值进行归一化处理,所述RGB值归一化处理公式包括:其中,x R、x G、x B分别为原有的所述预设分辨率的植物样本子图像对应的任一像素点的R、G、B通道的分量值,y R、y G、y B分别为与x R、x G、x B对应的像素点经过归一化处理后的R、G、B通道的分量值,x Rmax、x Gmax、x Bmax分别为原有的所述预设分辨率的植物样本子图像中在R、G、B通道上的最大分量值;所述样本集建立模块,具体用于按照第一预设次数和第二预设次数,分别对RGB值归一化处理后的所述预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本集。
- 根据权利要求12所述的装置,其特征在于,所述预设植物部位包括但不限于以下任意一种或其组合:植物根部、植物叶子、植物茎部。
- 根据权利要求13所述的装置,其特征在于,所述图像识别模块,具体包括:图像分辨率处理单元,用于在接收到的待识别的植物图像上按照所述预设植物部位进行图像截取,得到与所述预设植物部位对应的待识别的植物子图像;图像分块单元,用于对所述待识别的植物子图像进行分辨率预处理,以使所述待识别的植物子图像的分辨率与所述预设分辨率一致;图像RGB值归一化单元,用于按照所述RGB值归一化处理公式,对所述待识别的植物子图像的每个像素点的RGB值进行归一化处理;图像识别单元,用于将RGB值归一化处理后的所述待识别的植物子图像输入至所述训练好的模型中,得到与所述待识别的植物子图像对应的识别结果;识别结果计算单元,用于若所述预设植物部位包括一个,则将与所述待识别的植物子图像对应的识别结果作为所述待识别的植物图像的识别结果;以及若所述预设植物部位包括多个,则按照与多个所述预设植物部位对应的预设权重,对多个相应的所述识别结果进行加权求和计算,得到所述待识别的图像的识别结果。
- 一种计算机可读存储介质,其上存储有计算机可读指令,其特征在于,所述计算机可读指令被处理器执行时实现基于神经网络模型的植物图像的识别方法,包括:在植物样本图像上按照预设植物部位进行图像截取,得到与所述预设植物部位对应的植物样本子图像;对所述植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本子图像;按照第一预设次数和第二预设次数,分别对所述预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本集,其中,所述第一预设次数大于所述第二预设次数;利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模 型识别所述测试样本集的成功率大于或等于预设成功率;接收待识别的植物图像,并利用所述训练好的模型对所述待识别的植物图像的种类进行识别。
- 根据权利要求15所述的计算机可读存储介质,其特征在于,所述计算机可读指令被处理器执行时实现所述利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别所述测试样本集的成功率大于或等于预设成功率,具体包括:利用所述训练样本集,按照预设学习率对所述模型进行训练;若训练后的所述模型识别所述测试样本集的成功率小于所述预设成功率,则降低所述预设学习率,并按照降低后的预设学习率继续训练所述模型直至训练后的所述模型识别所述测试样本集的成功率大于或等于所述预设成功率为止。
- 根据权利要求15所述的计算机可读存储介质,其特征在于,所述计算机可读指令被处理器执行时实现所述利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别所述测试样本集的成功率大于或等于预设成功率,具体包括:利用所述训练样本集,按照预设学习率对所述模型进行训练;若训练后的所述模型识别所述测试样本集的成功率小于所述预设成功率,则降低所述预设学习率,并按照降低后的预设学习率继续训练所述模型直至训练后的所述模型识别所述测试样本集的成功率大于或等于所述预设成功率为止。
- 一种计算机设备,包括存储介质、处理器及存储在存储介质上并可在处理器上运行的计算机可读指令,其特征在于,所述处理器执行所述程序时实现基于神经网络模型的植物图像的识别方法,包括:在植物样本图像上按照预设植物部位进行图像截取,得到与所述预设植物部位对应的植物样本子图像;对所述植物样本子图像进行分辨率预处理,得到预设分辨率的植物样本子图像;按照第一预设次数和第二预设次数,分别对所述预设分辨率的植物样本子图像进行有放回的抽样,得到训练样本集和测试样本集,其中,所述第一预设次数大于所述第二预设次数;利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别所述测试样本集的成功率大于或等于预设成功率;接收待识别的植物图像,并利用所述训练好的模型对所述待识别的植物图像的种类进行识别。
- 根据权利要求18所述的计算机设备,其特征在于,所述处理器执行所述计算机 可读指令时实现所述利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别所述测试样本集的成功率大于或等于预设成功率,具体包括:利用所述训练样本集,按照预设学习率对所述模型进行训练;若训练后的所述模型识别所述测试样本集的成功率小于所述预设成功率,则降低所述预设学习率,并按照降低后的预设学习率继续训练所述模型直至训练后的所述模型识别所述测试样本集的成功率大于或等于所述预设成功率为止。
- 根据权利要求18所述的计算机设备,其特征在于,所述处理器执行所述计算机可读指令时实现所述利用所述训练样本集训练MobileNetV2轻量级卷积神经网络模型,直至训练好的模型识别所述测试样本集的成功率大于或等于预设成功率,具体包括:利用所述训练样本集,按照预设学习率对所述模型进行训练;若训练后的所述模型识别所述测试样本集的成功率小于所述预设成功率,则降低所述预设学习率,并按照降低后的预设学习率继续训练所述模型直至训练后的所述模型识别所述测试样本集的成功率大于或等于所述预设成功率为止。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910185076.9 | 2019-03-12 | ||
| CN201910185076.9A CN110070101B (zh) | 2019-03-12 | 2019-03-12 | 植物种类的识别方法及装置、存储介质、计算机设备 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020181817A1 true WO2020181817A1 (zh) | 2020-09-17 |
Family
ID=67366192
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/118395 Ceased WO2020181817A1 (zh) | 2019-03-12 | 2019-11-14 | 一种基于神经网络模型的植物图像的识别方法及装置 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN110070101B (zh) |
| WO (1) | WO2020181817A1 (zh) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113344035A (zh) * | 2021-05-17 | 2021-09-03 | 捷佳润科技集团股份有限公司 | 一种香蕉物候期监控模块及种植系统 |
| CN116310727A (zh) * | 2023-03-22 | 2023-06-23 | 国网浙江省电力有限公司超高压分公司 | 一种基于轻量级网络的电力设备的识别方法及相关装置 |
| CN117036828A (zh) * | 2023-09-19 | 2023-11-10 | 南方电网数字电网研究院有限公司 | 用于保护输电线路的速生树监测方法、装置、设备和介质 |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110070101B (zh) * | 2019-03-12 | 2024-05-14 | 平安科技(深圳)有限公司 | 植物种类的识别方法及装置、存储介质、计算机设备 |
| CN110487741A (zh) * | 2019-08-22 | 2019-11-22 | Oppo(重庆)智能科技有限公司 | 灌溉信息确定方法、装置及终端设备 |
| CN110852341B (zh) * | 2019-09-23 | 2024-12-13 | 平安科技(深圳)有限公司 | 基于深度学习的白术检测方法、及其相关设备 |
| US11138787B2 (en) * | 2019-11-25 | 2021-10-05 | Rockwell Collins, Inc. | Efficient transfer of dynamic 3D world model data |
| CN112906437B (zh) * | 2019-12-04 | 2023-12-26 | 中国农业科学院农业资源与农业区划研究所 | 一种燕麦植株识别的方法及装置 |
| CN112906698B (zh) * | 2019-12-04 | 2023-12-29 | 中国农业科学院农业资源与农业区划研究所 | 一种苜蓿植株识别的方法及装置 |
| CN111568195A (zh) * | 2020-02-29 | 2020-08-25 | 佛山市云米电器科技有限公司 | 冲泡饮物识别方法、设备及计算机可读存储介质 |
| CN113627216B (zh) * | 2020-05-07 | 2024-02-27 | 杭州睿琪软件有限公司 | 植物状态评估方法、系统及计算机可读存储介质 |
| CN113807132B (zh) * | 2020-06-12 | 2023-05-02 | 广州极飞科技股份有限公司 | 植物生长区域的灌水状态的识别方法、装置和存储介质 |
| CN111723736A (zh) * | 2020-06-19 | 2020-09-29 | 中国农业科学院农业信息研究所 | 果树花期监测方法、装置、计算机设备及存储介质 |
| CN112801039A (zh) * | 2021-03-03 | 2021-05-14 | 广西广播电视技术中心 | 一种基于改进MobilenetV2网络的劣质数字电视画面识别方法 |
| CN113313193A (zh) * | 2021-06-15 | 2021-08-27 | 杭州睿胜软件有限公司 | 植物图片识别方法、可读存储介质及电子设备 |
| CN113298180A (zh) * | 2021-06-15 | 2021-08-24 | 杭州睿胜软件有限公司 | 用于植物识别的方法和计算机系统 |
| CN117541225B (zh) * | 2023-11-11 | 2024-11-15 | 浙江浙勤城市服务科技有限公司 | 一种移植后树木的养护检测方法、系统及存储介质 |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103870816A (zh) * | 2014-03-26 | 2014-06-18 | 中国科学院寒区旱区环境与工程研究所 | 一种识别率高的植物识别方法及装置 |
| US20150036920A1 (en) * | 2013-07-31 | 2015-02-05 | Fujitsu Limited | Convolutional-neural-network-based classifier and classifying method and training methods for the same |
| CN107239514A (zh) * | 2017-05-19 | 2017-10-10 | 邓昌顺 | 一种基于卷积神经网络的植物识别方法及系统 |
| CN108765368A (zh) * | 2018-04-20 | 2018-11-06 | 平安科技(深圳)有限公司 | Mri病变位置检测方法、装置、计算机设备和存储介质 |
| CN108875620A (zh) * | 2018-06-06 | 2018-11-23 | 中国农业科学院深圳农业基因组研究所 | 入侵植物的监测方法及系统 |
| CN110070101A (zh) * | 2019-03-12 | 2019-07-30 | 平安科技(深圳)有限公司 | 植物种类的识别方法及装置、存储介质、计算机设备 |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AR104234A1 (es) * | 2016-04-12 | 2017-07-05 | Hernan Perez Roca Diego | Conjunto autónomo de dispositivos y método para la detección e identificación de especies vegetales en un cultivo agrícola para la aplicación de agroquímicos en forma selectiva |
| WO2018112783A1 (zh) * | 2016-12-21 | 2018-06-28 | 深圳前海达闼云端智能科技有限公司 | 图像识别方法及装置 |
| CN107346434A (zh) * | 2017-05-03 | 2017-11-14 | 上海大学 | 一种基于多特征及支持向量机的植物病虫害检测方法 |
| CN108256568B (zh) * | 2018-01-12 | 2021-10-01 | 宁夏智启连山科技有限公司 | 一种植物种类识别方法以及装置 |
-
2019
- 2019-03-12 CN CN201910185076.9A patent/CN110070101B/zh active Active
- 2019-11-14 WO PCT/CN2019/118395 patent/WO2020181817A1/zh not_active Ceased
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150036920A1 (en) * | 2013-07-31 | 2015-02-05 | Fujitsu Limited | Convolutional-neural-network-based classifier and classifying method and training methods for the same |
| CN103870816A (zh) * | 2014-03-26 | 2014-06-18 | 中国科学院寒区旱区环境与工程研究所 | 一种识别率高的植物识别方法及装置 |
| CN107239514A (zh) * | 2017-05-19 | 2017-10-10 | 邓昌顺 | 一种基于卷积神经网络的植物识别方法及系统 |
| CN108765368A (zh) * | 2018-04-20 | 2018-11-06 | 平安科技(深圳)有限公司 | Mri病变位置检测方法、装置、计算机设备和存储介质 |
| CN108875620A (zh) * | 2018-06-06 | 2018-11-23 | 中国农业科学院深圳农业基因组研究所 | 入侵植物的监测方法及系统 |
| CN110070101A (zh) * | 2019-03-12 | 2019-07-30 | 平安科技(深圳)有限公司 | 植物种类的识别方法及装置、存储介质、计算机设备 |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113344035A (zh) * | 2021-05-17 | 2021-09-03 | 捷佳润科技集团股份有限公司 | 一种香蕉物候期监控模块及种植系统 |
| CN116310727A (zh) * | 2023-03-22 | 2023-06-23 | 国网浙江省电力有限公司超高压分公司 | 一种基于轻量级网络的电力设备的识别方法及相关装置 |
| CN117036828A (zh) * | 2023-09-19 | 2023-11-10 | 南方电网数字电网研究院有限公司 | 用于保护输电线路的速生树监测方法、装置、设备和介质 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110070101B (zh) | 2024-05-14 |
| CN110070101A (zh) | 2019-07-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020181817A1 (zh) | 一种基于神经网络模型的植物图像的识别方法及装置 | |
| WO2020181706A1 (zh) | 植物种类的识别方法及装置 | |
| CN110163080B (zh) | 人脸关键点检测方法及装置、存储介质和电子设备 | |
| US11151723B2 (en) | Image segmentation method, apparatus, and fully convolutional network system | |
| CN109871845B (zh) | 证件图像提取方法及终端设备 | |
| CN112183449B (zh) | 一种驾驶员身份验证方法、装置、电子设备及存储介质 | |
| CN114005029A (zh) | 基于改进yolov5网络的广佛手病虫害识别方法及系统 | |
| WO2020258667A1 (zh) | 图像识别方法及装置、非易失性可读存储介质、计算机设备 | |
| CN110334585A (zh) | 表格识别方法、装置、计算机设备和存储介质 | |
| CN108717524A (zh) | 一种基于双摄手机和人工智能系统的手势识别系统及方法 | |
| CN108734673A (zh) | 去网纹系统训练方法、去网纹方法、装置、设备及介质 | |
| CN117372881A (zh) | 一种烟叶病虫害智能识别方法、介质及系统 | |
| CN107506688B (zh) | Harris角点检测影像金字塔掌纹ROI识别方法 | |
| CN110516676A (zh) | 一种基于图像处理的银行卡号识别系统 | |
| CN111860601B (zh) | 预测大型真菌种类的方法及装置 | |
| CN111860316A (zh) | 一种驾驶行为的识别方法、装置及存储介质 | |
| CN115761834A (zh) | 人脸识别的多任务混合模型和人脸识别方法 | |
| CN113436081A (zh) | 数据处理方法、图像增强方法及其模型训练方法 | |
| CN111914068B (zh) | 试题知识点的提取方法 | |
| CN116665092B (zh) | 基于ia-yolov7的污水悬浮物识别方法及系统 | |
| CN106250871A (zh) | 城市管理案件分类方法及装置 | |
| CN117292220A (zh) | 基于轻量化目标检测算法的叶片病害检测模型建立方法 | |
| CN116485775A (zh) | 电杆受损图像的识别方法和装置、电子设备和存储介质 | |
| CN117037256A (zh) | 一种基于轻量表情识别模型的课堂专注度检测方法及装置 | |
| CN111144407A (zh) | 一种目标检测方法、系统、装置及可读存储介质 |
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: 19919498 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS (EPO FORM 1205A DATED 03.11.2021) |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 19919498 Country of ref document: EP Kind code of ref document: A1 |
