Foreign matter identification method and system and photovoltaic inspection equipment

Abstract

The invention relates to the field of photovoltaic inspection, and discloses a foreign matter identification method and system and photovoltaic inspection equipment, and the method comprises the steps: obtaining a data set containing a foreign matter, carrying out the division of at least two dimensions of the foreign matter in the data set, and adding a foreign matter type label to each data subset obtained through division; constructing a foreign matter recognition model, and training the foreign matter recognition model by using the data set; and identifying a target foreign matter on the photovoltaic module based on the trained foreign matter identification model, and outputting a feature value of the target foreign matter corresponding to each foreign matter category label. Whether the target foreign matter exists in the image data or not is recognized, and the probability value of the target foreign matter corresponding to each foreign matter category label is obtained through calculation. According to the method and the device, the data set is divided in different dimensions, so that the data set has a generalization characteristic, the model has a basis for identifying any foreign matter, the difficulty of data set arrangement is reduced, the working performance of model construction and processing is improved, and the foreign matter identification range is also expanded.

Description

Technical Field

[0001] The present application relates to the field of photovoltaic inspection, and more specifically, to a foreign object identification method, system and photovoltaic inspection equipment. Background Art

[0002] Solar photovoltaic panels are an important part of modern clean energy technology. They provide power to homes, businesses and the grid by converting sunlight into electricity. However, the efficiency of photovoltaic panels is affected by many factors, one of which is foreign matter on the surface. When leaves, plastic waste, bird droppings or any other form of foreign matter appear on photovoltaic panels, they block direct sunlight and affect the ability of photovoltaic panels to absorb light energy.

[0003] If these foreign objects form a large-scale obstruction, they may also cause local hot spots on the photovoltaic panels, which may lead to fires and safety accidents. Traditional manual inspection and cleaning methods are inadequate when facing large-scale photovoltaic power plants. Due to the large number of photovoltaic panels, manually checking whether each panel is obstructed is not only time-consuming, but also increases labor costs. In addition, the efficiency and accuracy of manual inspections are difficult to guarantee, and obstructions may be missed during inspection. Summary of the invention

[0004] In order to solve the above technical problems, the present application provides a foreign object identification method, system and photovoltaic inspection equipment. By dividing a large number of photovoltaic component data sets of a photovoltaic power station into different dimensions, the data sets are generalized so that the model has the basis for identifying any foreign object. A large number of different types of foreign object features are not required in the data set, which improves the working performance of model construction and processing, and also increases the scope of foreign object identification.

[0005] Specifically, the technical solution of this application is as follows:

[0006] In a first aspect, the present application discloses a foreign body identification method, comprising:

[0007] Acquire a data set containing foreign matter, divide the features of the foreign matter in the data set into at least two dimensions, and add a foreign matter category label to each data subset obtained by the division;

[0008] Constructing a foreign body recognition model, and using the data set to train the foreign body recognition model;

[0009] Based on the trained foreign body recognition model, the collected image data is recognized and outputted, and the feature value of each foreign body category label corresponding to the image data is outputted;

[0010] The feature value is processed by a probability conversion function to identify whether a target foreign object exists in the image data, and a probability value of the target foreign object corresponding to each of the foreign object category labels is obtained.

[0011] In some embodiments, dividing the data set into at least two dimensions and adding a foreign body category label to each data subset obtained by the division includes:

[0012] Identify the color of the foreign object, divide the data set and add color labels;

[0013] Identify the shape of the foreign body, divide the data set and add shape labels;

[0014] Identifying the size of the foreign object relative to the photovoltaic module, dividing the data set and adding size tags;

[0015] It is identified whether the location of the foreign object is on the photovoltaic component, the data set is divided and a location tag is added.

[0016] In some embodiments, the foreign body recognition model includes a convolutional neural network and a multi-layer perceptron structure connected to each other.

[0017] In some embodiments, the method of recognizing the collected image data based on the trained foreign body recognition model and outputting the feature value of each foreign body category label corresponding to the image data comprises the following steps:

[0018] Inputting the image data into the convolutional neural network;

[0019] Extracting image features from the image data using the convolutional neural network to obtain image features;

[0020] The image features are classified through the multi-layer perceptron structure, and the feature values ​​of the image data corresponding to each foreign object category label are output.

[0021] In some embodiments, the method of classifying the image features through the multi-layer perceptron structure and outputting the feature value of each foreign body category label corresponding to the image data further includes:

[0022] By using the multi-layer perceptron structure, the feature values ​​are divided into a first type of feature values ​​with foreign body features and a second type of feature values ​​without foreign body features;

[0023] Specifically, based on the characteristic value of the location tag, whether the foreign object is within the range of the photovoltaic component is classified, and if so, the characteristic value is classified as a first-category characteristic value; otherwise, the characteristic value is classified as a second-category characteristic value;

[0024] Based on the characteristic value of the size label, it is classified whether the size of the foreign body is larger than the minimum set size. If so, the characteristic value is classified as a first type of characteristic value; otherwise, the characteristic value is classified as a second type of characteristic value.

[0025] In some embodiments, the processing of the feature value by a probability conversion function to identify whether a target foreign body exists in the image data includes:

[0026] Processing the first-category feature value and the second-category feature value through a probability conversion function to obtain probability values ​​of the probability of foreign matter and the probability of no foreign matter in the foreign matter category label corresponding to the image data;

[0027] If the probability value of the probability of foreign matter being present is greater than the first set probability, or the probability value of the probability of no foreign matter being present is less than the second set probability, it is considered that the target foreign matter exists in the image data.

[0028] In some embodiments, obtaining the probability value of the target foreign object corresponding to each of the foreign object category labels includes:

[0029] Processing the characteristic values ​​of the foreign body category labels in sequence through a probability conversion function;

[0030] Prioritizing processing of the position tag, determining the block position of the foreign object on the photovoltaic module, and obtaining a corresponding probability value;

[0031] Secondly, the size tag is processed to determine the size range of the foreign matter and obtain its corresponding probability value;

[0032] Re-identify the shape label to determine whether the foreign object has flake, strip, round, or irregular shape characteristics, and obtain corresponding probability values ​​respectively;

[0033] Finally, the color labels are processed to identify the color categories to which the foreign matter belongs, and the corresponding probability values ​​are obtained.

[0034] In some embodiments, the foreign body recognition method further includes using the verified image data to perform back-propagation training on the foreign body recognition model.

[0035] In some embodiments, the foreign body identification method further includes: controlling the identification strength of the corresponding foreign body by adjusting the feature weight for identifying the foreign body.

[0036] In a second aspect, the present application also summarizes a foreign body identification system, including a memory, a processor, and a computer program stored on the memory, wherein the processor executes the computer program to implement the steps of a foreign body identification method described in any one of the above embodiments.

[0037] In a third aspect, the present application further discloses a photovoltaic inspection device, comprising at least a foreign object identification system as described in any one of the above-mentioned embodiments.

[0038] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0039] 1. The data set constructed by this application can reduce the processing difficulty of the deep learning model and improve the accuracy of the deep learning model in processing a large number of foreign objects. Through this application, the data set is divided into different dimensions, so that it has generalized characteristics, which can enable the model to have the basis for identifying any foreign object. The difficulty and workload of data set collection are reduced.

[0040] 2. The foreign object recognition model provided in this application is trained based on the data set provided in this application. It does not require semantic segmentation and detection of the input image data. It only needs to be able to achieve classification tasks to recognize different types of foreign object features, thereby improving the performance of model construction and processing, and also increasing the scope of foreign object recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The preferred implementation scheme will be described below in a clear and understandable manner with reference to the accompanying drawings to further illustrate the above-mentioned characteristics, technical features, advantages and implementation methods of the present application.

[0042] Figure 1 A flowchart of a method for identifying foreign matter according to an embodiment of the present invention;

[0043] Figure 2 A flowchart of another embodiment of a foreign body identification method of the present application;

[0044] Figure 3 A schematic diagram of the structure of a foreign body recognition model in an embodiment provided in the present application;

[0045] Figure 4 This is a schematic diagram of the structure of a foreign body recognition model in another embodiment provided in the present application. DETAILED DESCRIPTION

[0046] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0047] It should be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections.

[0048] In order to simplify the drawings, only the parts related to the invention are schematically shown in each figure, and they do not represent the actual structure of the product. In addition, in order to simplify the drawings and facilitate understanding, in some figures, only one of the parts with the same structure or function is schematically drawn or marked. In this article, "one" not only means "only one", but also means "more than one".

[0049] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0050] In this document, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0051] In a specific implementation, the terminal device described in the embodiments of the present application includes, but is not limited to, other portable devices such as mobile phones, laptop computers, tutoring machines, or tablet computers with touch-sensitive surfaces (e.g., touch screen displays and / or touch pads). It should also be understood that in some embodiments, the terminal device is not a portable communication device, but a desktop computer with a touch-sensitive surface (e.g., touch screen displays and / or touch pads).

[0052] In addition, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0053] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the specific implementation methods of the present application will be described below with reference to the accompanying drawings. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings and other implementation methods can be obtained based on these drawings without creative work.

[0054] In the existing technology, in order to improve the maintenance efficiency of photovoltaic power generation systems and reduce operating costs, the identification of foreign objects on photovoltaic modules tends to be more automated and intelligent. For example, using drones for inspections, or developing intelligent algorithms to automatically detect foreign objects on photovoltaic panels through image recognition technology. These technologies can not only quickly cover large areas of photovoltaic panels, but also provide more accurate and consistent inspection results, thereby ensuring that photovoltaic power generation systems can operate at the highest efficiency. These solutions are usually inseparable from suitable artificial intelligence (deep learning) models.

[0055] At present, there are several prominent problems in using artificial intelligence technology for automatic identification. The first is that there is no ready-made model that can be used to identify all foreign objects on photovoltaic modules. The second is that there are many types of foreign objects with different characteristics. If the common model training method is used, it is necessary to collect multiple images for each type of foreign object as training set images. The construction of the data set is cumbersome and difficult, which increases the cost of model construction.

[0056] In view of the above problems, the present application provides a foreign body identification method, which improves the scope of foreign body identification of the model by dividing the data set into different dimensions. Based on the types of division, almost all types of foreign bodies can be recognized. Figure 1 An embodiment of a foreign body identification method provided by the present application includes the following steps:

[0057] S100, obtaining a data set containing foreign matter, dividing the features of the foreign matter in the data set into at least two dimensions, and adding a foreign matter category label to each data subset obtained by the division.

[0058] S200, constructing a foreign body recognition model, and using the data set to train the foreign body recognition model.

[0059] S300, identifying the collected image data based on the trained foreign object recognition model, and outputting the feature value of each foreign object category label corresponding to the image data.

[0060] S400, processing the feature value through a probability conversion function to identify whether a target foreign object exists in the image data, and obtaining a probability value of the target foreign object corresponding to each of the foreign object category labels.

[0061] In one implementation of this embodiment, the data set is required to be considered from different dimensions and have generalized characteristics, that is, characteristics that can be generalized to any foreign object, so that the foreign object recognition obtained by training can have a basis for recognizing any foreign object. Therefore, when creating a deep learning data set, it is divided according to the following dimensions, including:

[0062] 1. Identify the color of the foreign matter, divide the data set and add color labels; specifically, divide from the dimension of color: 9 common colors can be used, including: red, orange, yellow, green, blue, indigo, purple, white and black.

[0063] 2. Identify the shape of the foreign object, divide the data set and add shape labels; specifically, it can be divided into 7 categories: cylinder, cube, cuboid, sheet, sphere, strip and irregular shape. Sheets refer to thin foreign objects such as leaves and paper, strips refer to ropes, wire harnesses and branches, and irregular shapes refer to foreign objects that do not belong to the previous 6 shapes, including gloves, bananas, etc.

[0064] 3. Identify the size of the foreign object relative to the photovoltaic module, divide the data set and add size labels; specifically, divide it from the dimension of size: it can be divided into 6 types: 1, 2, 4, 16, 72, and 144. More preferably, the photovoltaic panel can be divided into units, and a photovoltaic panel can be pre-divided into 144 cells in total, which can be divided based on length or width: 12×12, 8×18, 6×24 and other combinations. 1 means that the area of ​​the foreign object is close to 1 cell of the photovoltaic panel, 16 means that the area of ​​the foreign object is 16 cells, and so on. 72 means half of the area of ​​the photovoltaic panel. When the actual size of the object is less than 1 cell, it is processed as 1 cell; when the object spans multiple cells, it is processed as multiple cells. Of course, in actual applications, the size of the division size can be selected independently.

[0065] 4. Identify whether the location of the foreign object is on the photovoltaic component, divide the data set and add a location tag. Specifically, it can be divided according to the location of the foreign object: photovoltaic panels are usually rectangular, so foreign objects will appear on the long side or short side of the photovoltaic panel. Then, combining the two relationships inside the photovoltaic panel and outside the photovoltaic panel, 4 combinations can be obtained. Including: inside the long side, outside the long side, inside the short side, and outside the short side. Among them, foreign objects inside the long side and inside the short side can be regarded as the location of the foreign object above the photovoltaic panel. Foreign objects outside the long side and outside the short side can be regarded as the foreign object not being on the photovoltaic panel, for example, on the ground. In some embodiments, only foreign objects on the photovoltaic panel can be used as detection targets, and those not on the photovoltaic panel are not regarded as detection targets.

[0066] In this embodiment, after the above four dimensions are combined, 9*7*6*4=1512 combinations are obtained. Based on these four dimensions, the data set can be divided into 1512 data subsets, and corresponding category labels are added. These combinations can basically represent all types of foreign matter.

[0067] Preferably, in other embodiments, the data set can be divided by other dimensions, such as texture, material, volume, etc. Through this application, the data set is divided into different dimensions, so that it has generalized characteristics, which can enable the model to have the basis for identifying any foreign objects. The generalization ability of a machine learning model refers to the performance of the model when processing unseen data. This enables the model to make accurate predictions or classifications of new data that has not participated in training.

[0068] The generalization ability of a data set usually refers to the ability of the data set to represent the actual situation of the target task or problem, thereby supporting the training of a model with strong generalization ability. In this application, all possible foreign objects are classified by dividing the foreign objects. This method does not perform semantic segmentation or detection, but only uses the classification task of deep learning to identify all foreign objects. This reduces the training cost of the model.

[0069] The foreign body recognition model is trained using the data set. First, the model itself must have strong generalization capabilities and must be compatible with the 1512 combinations in the first step above. Therefore, when implementing, a convolutional neural network structure is selected as the basis, its output features are extracted, and then a new classifier (multilayer perceptron) is created. Specifically, another embodiment of a foreign body recognition method of the present application is as shown in the attached specification. Figure 2 As shown, based on an embodiment of the above method, step S300 specifically includes the following steps:

[0070] S310, inputting the image data into the convolutional neural network;

[0071] S320, extracting image features from the image data by using the convolutional neural network to obtain image features;

[0072] S330, classifying the image features through the multi-layer perceptron structure, and outputting the feature value of each foreign object category label corresponding to the image data.

[0073] Specifically, in this embodiment, the RegNet network is used as the basis, and the MLP (multilayer perceptron) structure is used as the classifier. Of course, other network models are used as the basis, such as EfficientNet, ResNet, etc.; other multilayer perceptron structures are used as classifiers, such as feedforward neural networks, deep residual networks, etc., which can also achieve the technical effects that can be achieved in this embodiment.

[0074] In this embodiment, the schematic diagram of the foreign body recognition model is shown in FIG. Figure 3 As shown. Among them, the nodes of the I layer are the nodes of the convolutional neural network output features, and the nodes of the II layer are the feature value nodes corresponding to the combination of multiple foreign body category labels. If you need to directly obtain the probability of each of the 1512 combinations above, and then combine the hard-coded method to determine whether there is a foreign body, you can modify the structure of the multilayer perceptron so that the II layer nodes of the multilayer perceptron (classifier) ​​directly output 1512 values, corresponding to the 1512 combinations of the data set. Then use the probability conversion function to convert the numerical value into a probability. Finally, the model is continuously trained so that these probability values ​​can correctly represent the probabilities of each category. In this embodiment, the sigmoid function is used as the probability conversion function. Of course, the use of other existing probability conversion functions, such as: Logistic function, Softmax function, etc., can also achieve similar technical effects as the present application.

[0075] After the probability values ​​of each foreign body category label corresponding to the image data are output, it is finally manually encoded and the presence of foreign bodies is judged based on these 1512 probability values.

[0076] In another implementation of this embodiment, as shown in the attached specification Figure 2 As shown, based on an embodiment of the above method, the step S300 further includes: step S340, dividing the feature values ​​into a first type of feature values ​​with foreign body features and a second type of feature values ​​without foreign body features through the multi-layer perceptron structure.

[0077] Specifically, it includes: based on the characteristic value of the location tag, classifying whether the foreign object is within the range of the photovoltaic component, and if so, classifying the characteristic value into a first type of characteristic value; otherwise, classifying the characteristic value into a second type of characteristic value.

[0078] Based on the characteristic value of the size label, it is classified whether the size of the foreign body is larger than the minimum set size. If so, the characteristic value is classified as a first-class characteristic value; otherwise, the characteristic value is classified as a second-class characteristic value.

[0079] Optionally, the schematic diagram of the foreign body recognition model in this embodiment is referenced Figure 4As shown; wherein, the nodes of layer I are nodes of the output features of the convolutional neural network, and the nodes of layer II are feature value nodes corresponding to the combination of multiple foreign body category labels. The nodes of layer III include first-class feature value nodes and second-class feature value nodes. This embodiment adds a layer to the MLP structure described in the above embodiment to classify the 1512 types of feature data output, and finally divides the feature values ​​into two categories. The two output categories represent the two situations of "foreign body" and "no foreign body".

[0080] In this embodiment, the I-layer nodes, II-layer nodes, and III-layer nodes are only used to distinguish and describe; they are not used to indicate the hierarchical relationship and number of layers of nodes. For example, there may be several layers between the I-layer nodes and the II-layer nodes. Figure 3 and Figure 4 Not drawn in.

[0081] More preferably, after step S340, it also includes: S410: Processing the first type of feature value and the second type of feature value through a probability conversion function to obtain probability values ​​of the probability of foreign matter and the probability of no foreign matter in the image data. Specifically, the sigmoid function is used to convert the values ​​of the two nodes into probabilities. Finally, the model is continuously trained so that these probability values ​​can correctly represent the probability of foreign matter. If the probability value is lower than a preset threshold, it is considered that there is no foreign matter. On the contrary, if a probability value is higher than a preset threshold, it is considered that the model has identified a foreign matter.

[0082] Furthermore, if the probability value of the probability of foreign matter being present is greater than the first set probability, or the probability value of the probability of no foreign matter being present is less than the second set probability, it is considered that the target foreign matter is present in the image data.

[0083] On the contrary, if the probability value of the probability of foreign matter is less than the first set probability, or the probability value of the probability of no foreign matter is greater than the second set probability, it is considered that there is no target foreign matter in the image data; or there is a non-target foreign matter.

[0084] In some implementations, the magnitudes of the first set probability and the second set probability may be set, the first set probability is equal to the second set probability, or the first set probability is not equal to the second set probability.

[0085] After determining that there is a target foreign body, the next step is to process the feature values ​​of the foreign body category label in sequence through a probability conversion function.

[0086] The position tag is processed first, the block position of the foreign object on the photovoltaic component is determined, and the corresponding probability value is obtained.

[0087] Secondly, the size tag is processed to determine the size range of the foreign matter and obtain its corresponding probability value.

[0088] The shape label is identified again to determine whether the foreign matter has sheet, strip, round, or irregular shape characteristics, and the corresponding probability values ​​are obtained respectively.

[0089] Finally, the color labels are processed to identify the color categories to which the foreign matter belongs, and the corresponding probability values ​​are obtained.

[0090] The preferred final sorting output obtains the position, size, shape, color, etc. of the target foreign matter, so as to identify the target foreign matter and process it to improve the photovoltaic power generation efficiency.

[0091] Another embodiment of a foreign body recognition method of the present application, based on any one of the embodiments of the above method, further includes: using the verified image data to perform back propagation training on the foreign body recognition model. Specifically, the gradient of the loss function relative to the network parameters is calculated by the back propagation algorithm, the network parameters are updated, and the back propagation training is combined to achieve that the foreign body recognition model can gradually give the probability of each foreign body category.

[0092] More preferably, the recognition strength of the corresponding foreign object is controlled by adjusting the feature weights for identifying the foreign object. Specifically, if you want to enhance the recognition effect of a certain type of foreign object in future specific use, you can use different weights for its corresponding features to highlight the recognition effect of this type of foreign object.

[0093] Based on the same technical concept, the present application also discloses a foreign body identification system, which can be used to implement any of the above foreign body identification methods. Specifically, an embodiment of a foreign body identification system of the present application includes:

[0094] A data set acquisition module, used to acquire a data set containing foreign matter, divide the features of the foreign matter in the data set into at least two dimensions, and add a foreign matter category label to each data subset obtained by the division;

[0095] A model training module, used to construct a foreign body recognition model and train the foreign body recognition model using the data set;

[0096] A feature output module, used to identify the collected image data based on the trained foreign body recognition model, and output the feature value of each foreign body category label corresponding to the image data;

[0097] The probability conversion module is used to process the feature value through a probability conversion function to obtain the probability value of the target foreign object corresponding to each of the foreign object category labels; thereby identifying whether the target foreign object exists in the image data and the category probability of the foreign object.

[0098] Based on the same concept, the present application also discloses a photovoltaic inspection device, which at least includes a foreign body identification system described in the above embodiment. The foreign body identification method, system and photovoltaic inspection device of the present application have the same technical concept, and the technical details of the embodiments of the three are applicable to each other. To reduce repetition, they will not be repeated here.

[0099] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the division of the above-mentioned program modules is used as an example for illustration. In actual applications, the above-mentioned functions can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program units or modules to complete all or part of the functions described above. The program modules in the embodiment can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into a processing unit, and the above-mentioned integrated unit can be implemented in the form of hardware or in the form of software program units. In addition, the specific names of the program modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application.

[0100] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0101] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A foreign body identification method, characterized in that: Used for photovoltaic inspection, including the following steps: Acquire a data set containing foreign matter, divide the features of the foreign matter in the data set into at least two dimensions, and add a foreign matter category label to each data subset obtained by the division; Constructing a foreign body recognition model, and using the data set to train the foreign body recognition model; Based on the trained foreign body recognition model, the collected image data is recognized and outputted, and the feature value of each foreign body category label corresponding to the image data is outputted; The feature value is processed by a probability conversion function to identify whether a target foreign object exists in the image data, and a probability value of the target foreign object corresponding to each of the foreign object category labels is obtained.

2. A foreign body identification method as claimed in claim 1, characterized in that: The step of dividing the data set into at least two dimensions and adding a foreign body category label to each data subset obtained by the division includes: Identify the color of the foreign object, divide the data set and add color labels; Identify the shape of the foreign body, divide the data set and add shape labels; Identifying the size of the foreign object relative to the photovoltaic module, dividing the data set and adding size tags; It is identified whether the location of the foreign object is on the photovoltaic component, the data set is divided and a location tag is added.

3. A foreign body identification method as claimed in claim 2, characterized in that: The foreign body recognition model includes a convolutional neural network and a multi-layer perceptron structure connected to each other; The method of recognizing the collected image data based on the trained foreign body recognition model and outputting the feature value of each foreign body category label corresponding to the image data comprises the following steps: Inputting the image data into the convolutional neural network; Extracting image features from the image data using the convolutional neural network to obtain image features; The image features are classified through the multi-layer perceptron structure, and the feature values ​​of the image data corresponding to each foreign object category label are output.

4. A foreign body identification method as claimed in claim 3, characterized in that: The method of classifying the image features through the multi-layer perceptron structure and outputting the feature value of each foreign body category label corresponding to the image data also includes: By using the multi-layer perceptron structure, the feature values ​​are divided into a first type of feature values ​​with foreign body features and a second type of feature values ​​without foreign body features; Specifically, it includes: based on the characteristic value of the location tag, classifying whether the foreign object is within the range of the photovoltaic component, and if so, classifying the characteristic value as a first-class characteristic value; if not, classifying the characteristic value as a second-class characteristic value; Based on the characteristic value of the size label, it is classified whether the size of the foreign body is larger than the minimum set size. If so, the characteristic value is classified as a first type of characteristic value; otherwise, the characteristic value is classified as a second type of characteristic value.

5. A foreign body identification method as claimed in claim 4, characterized in that: The step of processing the characteristic value by a probability conversion function to identify whether there is a target foreign object in the image data includes: Processing the first type of feature value and the second type of feature value through the probability conversion function to obtain probability values ​​of the probability of foreign matter and the probability of no foreign matter in the foreign matter category label corresponding to the image data; If the probability value of the probability of foreign matter being present is greater than the first set probability, or the probability value of the probability of no foreign matter being present is less than the second set probability, it is considered that the target foreign matter exists in the image data.

6. A foreign body identification method as claimed in claim 4, characterized in that: The obtaining of the probability value of the target foreign object corresponding to each of the foreign object category labels includes: Processing the characteristic values ​​of the foreign body category labels in sequence through a probability conversion function; Prioritizing processing of the position tag, determining the block position of the foreign object on the photovoltaic module, and obtaining a corresponding probability value; Secondly, the size tag is processed to determine the size range of the foreign matter and obtain its corresponding probability value; Re-identify the shape label to determine whether the foreign object has flake, strip, round, or irregular shape characteristics, and obtain corresponding probability values ​​respectively; Finally, the color labels are processed to identify the color categories to which the foreign matter belongs, and the corresponding probability values ​​are obtained.

7. A foreign body identification method according to any one of claims 1 to 6, characterized in that: Also includes: The verified image data is used to perform back-propagation training on the foreign object recognition model.

8. A foreign body identification method as claimed in claim 7, characterized in that: Also includes: By adjusting the feature weights for identifying the foreign objects, the recognition strength of the corresponding foreign objects is controlled.

9. A foreign body identification system, comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of a foreign body identification method as described in any one of claims 1-8.

10. A photovoltaic inspection device, characterized in that: At least includes a foreign body recognition system as described in claim 9.

Images