A Photovoltaic Equipment Detection Method and System Based on Machine Vision

Abstract

A photovoltaic device detection method and system based on machine vision provided by the present invention relate to the technical field of machine vision. In the present invention, first, an equipment relationship map is determined based on at least one cell image corresponding to a photovoltaic cell; secondly, a cell state vector corresponding to the target photovoltaic cell is determined based on the cell relationship map corresponding to the target photovoltaic cell in the equipment relationship map; then, based on the matching relationship between the cell state vector corresponding to the target photovoltaic cell and the cell state vectors corresponding to at least one photovoltaic cell state information to be confirmed, at least one photovoltaic cell state information corresponding to the target photovoltaic cell is analyzed. Based on the above method, the problem of relatively low reliability of photovoltaic device detection existing in the prior art can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of machine vision, and in particular, to a method and system for detecting photovoltaic devices based on machine vision. Background Art

[0002] The detection of photovoltaic devices is an important guarantee for the stable output of photovoltaic devices. For example, when defects (such as cracks, poor connections, deformations, etc.) are detected in photovoltaic devices, the photovoltaic devices are updated and maintained in a timely manner. Among them, traditional detection technologies generally rely on manual detection, which has problems such as high labor costs and low detection efficiency, and is also prone to problems such as missed detections. In the prior art, detection is generally based on image recognition technology. However, through research, it is found that when detecting based on conventional image recognition technology, there are problems with low reliability. Summary of the Invention

[0003] In view of this, an object of the present invention is to provide a method and system for detecting photovoltaic devices based on machine vision to improve the relatively low reliability of photovoltaic device detection in the prior art.

[0004] To achieve the above object, the embodiments of the present invention adopt the following technical solutions:

[0005] A method for detecting a photovoltaic device based on machine vision, comprising:

[0006] Determining a device relationship graph based on at least one cell image corresponding to a photovoltaic cell, wherein the device relationship graph includes a cell member corresponding to the photovoltaic cell, an image member corresponding to the cell image, and a relationship representation line, and the relationship representation line is used to reflect the correlation between the photovoltaic cell corresponding to the cell member and the cell image corresponding to the image member;

[0007] Determining a cell state vector corresponding to the target photovoltaic cell based on the cell relationship graph corresponding to the target photovoltaic cell in the device relationship graph, wherein the cell relationship graph corresponding to the target photovoltaic cell includes a target cell member corresponding to the target photovoltaic cell and at least one image member connected by the target cell member through the relationship representation line, and the cell state vector is used to reflect the state semantics of the target photovoltaic cell;

[0008] Analyzing at least one photovoltaic cell state information corresponding to the target photovoltaic cell based on the matching relationship between the cell state vector corresponding to the target photovoltaic cell and the cell state vectors corresponding to at least one photovoltaic cell state information to be confirmed, wherein the cell state vector corresponding to the photovoltaic cell state information is used to reflect the defect state semantics of the photovoltaic cell state information.

[0009] In some preferred embodiments, in the above-mentioned photovoltaic device detection method based on machine vision, the step of determining the cell state vector corresponding to the target photovoltaic cell based on the cell relationship graph corresponding to the target photovoltaic cell in the device relationship graph includes:

[0010] Based on at least one image member connected to the target cell member through the relationship representation line in the cell relationship graph, an image member set is determined;

[0011] Determine the cell state vector corresponding to the image member set, where the cell state vector corresponding to the image member set is used to reflect the state semantics of at least one cell image having a correlation relationship with the target photovoltaic cell;

[0012] Based on the cell state vector corresponding to the image member set, determine the cell state vector corresponding to the target photovoltaic cell.

[0013] In some preferred embodiments, in the above-mentioned photovoltaic device detection method based on machine vision, the step of determining an image member set based on at least one image member having the relationship representation line with the target cell member in the cell relationship graph includes:

[0014] For each image member among at least one image member connected to the target cell member through the relationship representation line, based on the image acquisition time configured on the relationship representation line between the image member and the target cell member, obtain the sequence relationship of the image members in the image member set, where the cell image corresponding to the image member refers to the historical cell image formed by image acquisition of the photovoltaic cell corresponding to the target cell member in history, and this historical cell image is used as the member attribute data of the image member in the device relationship graph. The member attribute data of the target cell member in the device relationship graph includes the current cell image formed by current image acquisition of the corresponding photovoltaic cell;

[0015] According to the corresponding sequence relationship, combine at least one image member connected to the target cell member through the relationship representation line to form the corresponding image member set.

[0016] In some preferred embodiments, in the above-mentioned photovoltaic device detection method based on machine vision, the step of determining the cell state vector corresponding to the image member set includes:

[0017] Obtain the member attribute data corresponding to each of at least one image member included in the image member set, where the member attribute data corresponding to the image member at least includes the cell image corresponding to the image member;

[0018] Load the member attribute data corresponding to each of the at least one image member, so that the photovoltaic cell mining model obtains the member attribute data;

[0019] Use the feature space mapping unit included in the photovoltaic cell mining model to perform a feature space mapping operation on the member attribute data corresponding to each of the at least one image member, and output the photovoltaic image mapping vector corresponding to the image member set;

[0020] Use the deep semantic mining unit in the photovoltaic cell mining model to perform a deep semantic mining operation on the photovoltaic image mapping vector corresponding to the image member set, and output the cell state vector corresponding to the image member set.

[0021] In some preferred embodiments, in the above-mentioned photovoltaic device detection method based on machine vision, the cell relationship graph further includes at least one related cell member connected to the target cell member through a relationship representation line, and there is an electrical property-related relationship between the photovoltaic cell corresponding to the related cell member and the photovoltaic cell corresponding to the target cell member, and the related relationship includes series connection and / or parallel connection;

[0022] Among them, the step of determining the cell state vector corresponding to the target photovoltaic cell based on the cell state vector corresponding to the image member set includes:

[0023] Use the feature space mapping unit included in the photovoltaic cell mining model to perform a feature space mapping operation on the member attribute data corresponding to the target cell member corresponding to the target photovoltaic cell and the member attribute data corresponding to each of at least one related cell member connected to the target cell member through a relationship representation line, and output the photovoltaic image mapping vector corresponding to the target cell member and the photovoltaic image mapping vectors corresponding to each of the at least one related cell member;

[0024] Use the deep semantic mining unit in the photovoltaic cell mining model to perform a deep semantic mining operation on the photovoltaic image mapping vector corresponding to the target cell member and the photovoltaic image mapping vectors corresponding to each of the at least one related cell member, and output the cell state vector corresponding to the target cell member and the cell state vectors corresponding to each of the at least one related cell member;

[0025] Determine the cell state vector corresponding to the target photovoltaic cell based on the cell state vectors corresponding to the set of image members, the cell state vector corresponding to the target cell member, and the cell state vectors corresponding to each of the at least one related cell member.

[0026] In some preferred embodiments, in the above-mentioned photovoltaic device detection method based on machine vision, the step of determining the cell state vector corresponding to the target photovoltaic cell based on the cell state vectors corresponding to the set of image members, the cell state vector corresponding to the target cell member, and the cell state vectors corresponding to each of the at least one related cell member includes:

[0027] Perform an association mining operation on each local vector in the cell state vector corresponding to the set of image members according to the order relationship corresponding to each image member in the set of image members, and output a first association mining vector corresponding to the set of image members;

[0028] Perform a difference operation with adjacent vectors on each local vector in the cell state vector corresponding to the set of image members according to the order relationship corresponding to each image member in the set of image members, and output a corresponding set of local difference vectors;

[0029] Perform an association mining operation on each local difference vector in the set of local difference vectors according to the order relationship corresponding to each image member in the set of image members, and output a second association mining vector corresponding to the set of image members;

[0030] Perform a vector aggregation operation on the first association mining vector corresponding to the set of image members and the second association mining vector corresponding to the set of image members, and output an aggregated association mining vector corresponding to the set of image members;

[0031] Perform an association focusing mining operation on the cell state vector corresponding to the target cell member respectively based on the aggregated association mining vector corresponding to the set of image members and the cell state vectors corresponding to each of the at least one related cell member, and output the cell state vector corresponding to the target photovoltaic cell.

[0032] In some preferred embodiments, in the above-mentioned photovoltaic device detection method based on machine vision, the step of performing an association focusing mining operation on the cell state vector corresponding to the target cell member respectively based on the aggregated association mining vector corresponding to the set of image members and the cell state vectors corresponding to each of the at least one related cell member, and outputting the cell state vector corresponding to the target photovoltaic cell includes:

[0033] Respectively take the aggregated association mining vector corresponding to the set of the image members and the cell state vectors corresponding to each of the at least one relevant cell member as the vectors to be aggregated, so as to obtain a plurality of vectors to be aggregated;

[0034] For each of the vectors to be aggregated, perform an association analysis operation on the vector to be aggregated and the cell state vector corresponding to the target cell member, output the correlation characterization parameter corresponding to the vector to be aggregated, and based on the correlation characterization parameter, perform an update operation on the vector to be aggregated, and output the associated focused mining vector corresponding to the vector to be aggregated;

[0035] Determine the central vector of the plurality of associated focused mining vectors corresponding to the plurality of vectors to be aggregated, and perform a superposition operation on the central vector and the cell state vector corresponding to the target cell member, and output the cell state vector corresponding to the target photovoltaic cell.

[0036] In some preferred embodiments, in the above-mentioned photovoltaic device detection method based on machine vision, the step of analyzing at least one photovoltaic cell state information corresponding to the target photovoltaic cell based on the matching relationship between the cell state vector corresponding to the target photovoltaic cell and the cell state vectors corresponding to at least one photovoltaic cell state information to be confirmed includes:

[0037] Obtain the cell state vectors corresponding to at least one photovoltaic cell state information to be confirmed respectively;

[0038] For each of the at least one photovoltaic cell state information to be confirmed, determine the vector correlation parameter between the cell state vector corresponding to the target photovoltaic cell and the cell state vector corresponding to the photovoltaic cell state information;

[0039] Among the at least one photovoltaic cell state information to be confirmed, determine the photovoltaic cell state information to be confirmed whose vector correlation parameter is greater than a pre-determined reference vector correlation parameter, and determine this photovoltaic cell state information as the photovoltaic cell state information corresponding to the target photovoltaic cell.

[0040] In some preferred embodiments, in the above-mentioned photovoltaic device detection method based on machine vision, the obtaining the cell state vectors corresponding to at least one photovoltaic cell state information to be confirmed respectively includes:

[0041] For each photovoltaic cell state information in at least one photovoltaic cell state information to be confirmed, at least one photovoltaic cell state associated information corresponding to the photovoltaic cell state information is extended, wherein there is a related defect state semantics between the photovoltaic cell state associated information and the photovoltaic cell state information;

[0042] Using the first feature space projection unit included in the state information mining model, perform a feature space projection operation on at least one state information unit included in the photovoltaic cell state information, and output a state information projection vector corresponding to the photovoltaic cell state information;

[0043] Using the second feature space projection unit included in the state information mining model, perform a feature space projection operation on at least one state information unit included in each of the at least one photovoltaic cell state associated information, and output a state information projection vector corresponding to the photovoltaic cell state associated information;

[0044] Using the vector aggregation unit included in the state information mining model, perform a vector aggregation operation on the state information projection vector corresponding to the photovoltaic cell state information and the state information projection vector corresponding to the photovoltaic cell state associated information, and output a corresponding state information aggregation vector;

[0045] Using the deep semantic representation unit included in the state information mining model, perform a deep semantic representation operation on the state information aggregation vector, and output a cell state vector corresponding to the photovoltaic cell state information.

[0046] An embodiment of the present invention further provides a photovoltaic device detection system based on machine vision, including a processor and a memory, the memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-mentioned photovoltaic device detection method based on machine vision.

[0047] A method and system for detecting photovoltaic devices based on machine vision provided by an embodiment of the present invention can determine a device relationship map based on at least one cell image corresponding to a photovoltaic cell; secondly, determine a cell state vector corresponding to a target photovoltaic cell based on the cell relationship map corresponding to the target photovoltaic cell in the device relationship map; then, analyze at least one photovoltaic cell state information corresponding to the target photovoltaic cell based on the matching relationship between the cell state vector corresponding to the target photovoltaic cell and the cell state vectors corresponding to at least one photovoltaic cell state information to be confirmed. Based on the above content, since the cell relationship map corresponding to the target photovoltaic cell includes the target cell members corresponding to the target photovoltaic cell and at least one image member connected by the relationship representation lines of the target cell members, when determining the cell state vector based on the cell relationship map, not only the information of the target cell members is referred to, but also the information of the image members is referred to, making the semantic information included in the cell state vector richer. Therefore, when analyzing the photovoltaic cell state information based on the cell state vector, a more comprehensive analysis basis can be available, making the reliability of the analysis result higher, that is, the reliability of the obtained photovoltaic cell state information is higher. Therefore, the problem of relatively low reliability in detecting photovoltaic devices existing in the prior art can be improved.

[0048] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, makes a detailed description as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a structural block diagram of a system for detecting photovoltaic devices based on machine vision provided by an embodiment of the present invention.

[0050] Figure 2 It is a schematic flowchart of each step included in a method for detecting photovoltaic devices based on machine vision provided by an embodiment of the present invention.

[0051] Figure 3 It is a schematic diagram of a device relationship map provided by an embodiment of the present invention.

[0052] Figure 4 It is a flowchart of association mining provided by an embodiment of the present invention.

[0053] Figure 5 It is a flowchart of determining a local difference vector provided by an embodiment of the present invention.

[0054] Figure 6 It is a flowchart of fusing multiple vectors provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part rather than all of the embodiments of the present invention. Components of the embodiments of the present invention described and illustrated herein usually can be arranged and designed in various different configurations.

[0056] Therefore, the detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0057] As Figure 1 shown, the embodiments of the present invention provide a photovoltaic device detection system based on machine vision. Among them, the photovoltaic device detection system based on machine vision may include a memory and a processor. Specifically, the memory and the processor are directly or indirectly electrically connected to achieve data transmission or interaction. For example, they may be electrically connected through one or more communication buses or signal lines. The memory may store at least one software functional module (computer program) that can exist in the form of software or firmware. The processor may be used to execute the executable computer program stored in the memory, thereby implementing the photovoltaic device detection method based on machine vision provided by the embodiments of the present invention (as described later).

[0058] Optionally, the memory may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), a System on Chip (SoC), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0059] And, Figure 1 The structure shown is only schematic, and the machine vision-based photovoltaic device detection system may further include more or fewer components than those shown in Figure 1 or have a different configuration from that shown in Figure 1 For example, it may include a communication unit for information interaction with other devices. Among them, in an alternative example, the machine vision-based photovoltaic device detection system may be a server with data processing capabilities.

[0060] In combination with Figure 2 , the embodiment of the present invention further provides a machine vision-based photovoltaic device detection method, which can be applied to the above-mentioned machine vision-based photovoltaic device detection system. Among them, the method steps defined by the processes related to the machine vision-based photovoltaic device detection method can be implemented by the machine vision-based photovoltaic device detection system.

[0061] Next, the specific process shown in Figure 2 will be elaborated in detail.

[0062] Step S110, determine a device relationship graph based on the cell images corresponding to at least one photovoltaic cell.

[0063] In an embodiment of the present invention, the machine vision-based photovoltaic device detection system may determine a device relationship graph based on cell images corresponding to at least one photovoltaic cell. Wherein, the device relationship graph includes cell members corresponding to the photovoltaic cells, image members corresponding to the cell images, and relationship representation lines, and the relationship representation lines are used to reflect the correlation between the photovoltaic cells corresponding to the cell members and the cell images corresponding to the image members. For example, the cell images are formed by performing image acquisition operations on the corresponding photovoltaic cells. It should be noted that the solution provided in the embodiment of the present invention can be applied before the formal use of the photovoltaic device or during the formal use of the photovoltaic device. When it is applicable during the formal use, when performing image acquisition operations through image acquisition devices such as cameras, due to the presence of the encapsulation glass layer, additional light sources (such as LED lights) can be used to ensure uniform illumination, so as to improve the clarity of the image by providing sufficient light. In addition, industrial cameras with high definition can also be used. In addition, a photovoltaic cell can refer to a single photovoltaic cell in the conventional sense or a component formed by connecting multiple photovoltaic cells in series or in parallel. That is to say, in the embodiment of the present invention, the division granularity of the photovoltaic cells in the photovoltaic device is not limited and can be selected according to actual needs. Among them, the formed device relationship graph can be as shown in Figure 3 shown.

[0064] Step S120, based on the cell relationship graph corresponding to the target photovoltaic cell in the device relationship graph, determine the cell state vector corresponding to the target photovoltaic cell.

[0065] In an embodiment of the present invention, the machine vision-based photovoltaic device detection system may determine the cell state vector corresponding to the target photovoltaic cell based on the cell relationship graph corresponding to the target photovoltaic cell in the device relationship graph. Wherein, the cell relationship graph corresponding to the target photovoltaic cell includes the target cell member corresponding to the target photovoltaic cell and at least one image member (such as Figure 3 cell member A, image member 11, image member 12, image member 13, and image member 14 in) connected to the target cell member through the relationship representation line, and the cell state vector is used to reflect the state semantics of the target photovoltaic cell. That is to say, by performing semantic mining on the information in the cell relationship graph, the corresponding vector is obtained.

[0066] Step S130, based on the matching relationship between the cell state vector corresponding to the target photovoltaic cell and the cell state vectors corresponding to at least one photovoltaic cell state information to be confirmed, analyze at least one photovoltaic cell state information corresponding to the target photovoltaic cell.

[0067] In an embodiment of the present invention, the machine vision-based photovoltaic device detection system can analyze at least one piece of photovoltaic cell state information (such as the most matching piece of photovoltaic cell state information) corresponding to the target photovoltaic cell based on the matching relationship between the cell state vector corresponding to the target photovoltaic cell and the cell state vectors corresponding to at least one piece of photovoltaic cell state information to be confirmed (pre-determined state classification labels). Among them, the cell state vector corresponding to the photovoltaic cell state information is used to reflect the defect state semantics of the photovoltaic cell state information, such as cracks, poor connections, deformations, etc. Based on this, after each photovoltaic cell is used as the target photovoltaic cell for corresponding processing, the photovoltaic cell state information corresponding to each photovoltaic cell can be obtained.

[0068] Based on the above, since the cell relationship graph corresponding to the target photovoltaic cell includes the target cell members corresponding to the target photovoltaic cell and at least one image member connected by the relationship representation line by the target cell members, when determining the cell state vector based on the cell relationship graph, not only the information of the target cell members is referred to, but also the information of the image members is referred to, making the semantic information included in the cell state vector richer. Therefore, when analyzing the photovoltaic cell state information based on the cell state vector, a more comprehensive analysis basis can be available, making the reliability of the analysis result higher, that is, the reliability of the obtained photovoltaic cell state information is higher. Therefore, the problem of relatively low reliability in photovoltaic device detection existing in the prior art can be improved.

[0069] Regarding step S120, it should be noted that the specific method for determining the cell state vector corresponding to the target photovoltaic cell is not limited and can be selected according to actual needs.

[0070] For example, in an alternative embodiment, in order to ensure that the determined cell state vector has a high semantic representation ability, step S120 described above can further include step S121, step S122, and step S123, and the specific content of each step is described as follows.

[0071] Step S121, based on at least one image member connected to the target cell member by the relationship representation line in the cell relationship graph, determine an image member set.

[0072] In an embodiment of the present invention, after determining the cell relationship graph corresponding to the target photovoltaic cell in the device relationship graph, an image member set can be determined based on at least one image member connected to the target cell member through the relationship representation line in the cell relationship graph. That is to say, each image member having a relevant relationship with the target cell member is screened out to form a set, namely, the image member set is obtained.

[0073] Step S122, determine the cell state vector corresponding to the image member set.

[0074] In an embodiment of the present invention, after obtaining the image member set, the cell state vector corresponding to the image member set can be determined. Among them, the cell state vector corresponding to the image member set is used to reflect the state semantics of at least one cell image having a relevant relationship with the target photovoltaic cell. For example, when cell member A is used as the target cell member, the corresponding cell state vector can be used to reflect the state semantics of the cell images corresponding to image member 11, image member 12, image member 13, and image member 14.

[0075] Step S123, determine the cell state vector corresponding to the target photovoltaic cell based on the cell state vector corresponding to the image member set.

[0076] In an embodiment of the present invention, after obtaining the cell state vector corresponding to the image member set, the cell state vector corresponding to the target photovoltaic cell can be determined based on the cell state vector corresponding to the image member set. That is to say, the cell state vector corresponding to the target photovoltaic cell at least includes the semantic information in the cell state vector corresponding to the image member set.

[0077] It can be understood that in the above step S121, the specific method for determining the image member set is not limited and can be selected according to actual needs. For example, in an alternative embodiment, for the convenience of subsequent semantic mining, the above step S121 may include:

[0078] First, for each of at least one image member connected to the target cell member through a relationship representation line, based on the image acquisition time configured on the relationship representation line between the image member and the target cell member, the order relationship of the image members in the image member set can be obtained. Here, the cell image corresponding to the image member refers to a historical cell image formed by acquiring an image of the photovoltaic cell corresponding to the target cell member in history. This historical cell image serves as the member attribute data that the image member has in the device relationship graph. The member attribute data that the target cell member has in the device relationship graph includes the current cell image formed by currently acquiring an image of the corresponding photovoltaic cell. It should be noted that during the formal use of photovoltaic equipment, the photovoltaic cell can be regularly imaged by an image acquisition device. Thus, for a photovoltaic cell, at least one frame of cell image can be formed, that is, the number of image members connected to the corresponding cell member is at least one (one frame of cell image serves as one image member).

[0079] Second, according to the corresponding order relationship, at least one image member connected to the target cell member through a relationship representation line can be combined to form a corresponding image member set. Exemplarily, the image members can be sorted and combined according to the order relationship of the image acquisition time from early to late to form a corresponding image member set. Based on this, since the image acquisition time of the cell image is carried in the image member set, the state of the photovoltaic cell changing over time is characterized.

[0080] It can be understood that in step S122 above, the specific method for determining the cell state vector corresponding to the image member set is not limited and can be selected according to actual needs. For example, in an alternative embodiment, in order to ensure the semantic representation ability of the mined cell state vector, step S122 above may include:

[0081] First, the member attribute data corresponding to each of at least one image member included in the image member set can be obtained. Here, the member attribute data corresponding to the image member at least includes the cell image corresponding to the image member. For example, the cell image can be directly used as the corresponding member attribute data.

[0082] Second, the member attribute data corresponding to each of the at least one image member is loaded, so that the photovoltaic cell mining model can obtain the member attribute data.

[0083] Then, the feature space mapping unit included in the photovoltaic cell mining model can be used to perform a feature space mapping operation on the member attribute data corresponding to each of the at least one image member, and output a photovoltaic image mapping vector corresponding to the image member set. It should be noted that the feature space mapping unit can be a convolutional network, which is used to perform convolutional processing on the cell image to implement the feature space mapping operation, so as to obtain the photovoltaic image mapping vector corresponding to the image member set. Specifically, after obtaining the convolutional vectors corresponding to each of the cell images, the convolutional vectors can be combined, such as concatenation, to form the photovoltaic image mapping vector corresponding to the image member set;

[0084] Finally, the deep semantic mining unit in the photovoltaic cell mining model can be used to perform a deep semantic mining operation on the photovoltaic image mapping vector corresponding to the image member set, and output a cell state vector corresponding to the image member set. That is to say, the photovoltaic image mapping vector can be further mined to obtain the deep semantic information therein, that is, the cell state vector corresponding to the image member set. Specifically, in an alternative embodiment, the photovoltaic image mapping vector corresponding to the image member set includes the convolutional vectors corresponding to each of the cell images. Thus, for each convolutional vector, self-attention processing can be performed on the convolutional vector (that is to say, the deep semantic mining unit can be a self-attention network) to obtain the corresponding local vector. Finally, the local vectors are combined together to form the cell state vector corresponding to the image member set.

[0085] It can be understood that in the above step S123, the specific manner of determining the cell state vector corresponding to the target photovoltaic cell is not limited and can be selected according to actual needs. For example, in an alternative embodiment, the cell relationship graph further includes at least one related cell member connected to the target cell member through a relationship representation line (for example, when the cell member B in Figure 3 is used as the target cell member, the related cell members include the photovoltaic cell member A and the photovoltaic cell member C), and there is an electrical property-related relationship between the photovoltaic cells corresponding to the related cell members and the photovoltaic cell corresponding to the target cell member. The related relationship includes series connection and / or parallel connection. Based on this, in order to further ensure the semantic representation ability of the mined cell state vector, the above step S123 may include step S123a, step S123b, and step S123c. The specific content of each step is described as follows.

[0086] Step S123a: Using the feature space mapping unit included in the photovoltaic cell mining model, perform feature space mapping operations on the member attribute data corresponding to the target cell member corresponding to the target photovoltaic cell and the member attribute data corresponding to each of at least one related cell member connected to the target cell member through a relationship representation line, and output the photovoltaic image mapping vector corresponding to the target cell member and the photovoltaic image mapping vector corresponding to each of the at least one related cell member.

[0087] In an embodiment of the present invention, the feature space mapping unit included in the photovoltaic cell mining model can be used to perform feature space mapping operations on the member attribute data corresponding to the target cell member corresponding to the target photovoltaic cell and the member attribute data corresponding to each of at least one related cell member connected to the target cell member through a relationship representation line, and output the photovoltaic image mapping vector corresponding to the target cell member and the photovoltaic image mapping vector corresponding to each of the at least one related cell member. That is, the member attribute data (cell image) corresponding to the target cell member can be convolved to obtain the corresponding photovoltaic image mapping vector, and the member attribute data corresponding to the related cell member can be convolved to obtain the corresponding photovoltaic image mapping vector.

[0088] Step S123b: Using the deep semantic mining unit in the photovoltaic cell mining model, perform deep semantic mining operations on the photovoltaic image mapping vector corresponding to the target cell member and the photovoltaic image mapping vector corresponding to each of the at least one related cell member, and output the cell state vector corresponding to the target cell member and the cell state vector corresponding to each of the at least one related cell member.

[0089] In an embodiment of the present invention, after obtaining the corresponding photovoltaic image mapping vector, the deep semantic mining unit in the photovoltaic cell mining model can be used to perform deep semantic mining operations on the photovoltaic image mapping vector corresponding to the target cell member and the photovoltaic image mapping vector corresponding to each of the at least one related cell member, and output the cell state vector corresponding to the target cell member and the cell state vector corresponding to each of the at least one related cell member. That is, the self-attention processing can be performed on the photovoltaic image mapping vector corresponding to the target cell member to obtain the corresponding cell state vector, and the self-attention processing can be performed on the photovoltaic image mapping vector corresponding to the related cell member to obtain the corresponding cell state vector.

[0090] Step S123c: Determine the cell state vector corresponding to the target photovoltaic cell based on the cell state vectors corresponding to the image member set, the cell state vector corresponding to the target cell member, and the cell state vectors corresponding to each of the at least one associated cell member.

[0091] In an embodiment of the present invention, after obtaining the corresponding cell state vectors, the cell state vector corresponding to the target photovoltaic cell can be determined based on the cell state vectors corresponding to the image member set, the cell state vector corresponding to the target cell member, and the cell state vectors corresponding to each of the at least one associated cell member. Based on this, the cell state vector corresponding to the target photovoltaic cell carries historical image semantics (i.e., the cell state vector corresponding to the image member set), current image semantics (i.e., the cell state vector corresponding to the target cell member), and current image semantics of the associated photovoltaic cells (i.e., the cell state vectors corresponding to each of the at least one associated cell member), enabling better semantic representation ability for subsequent state determination.

[0092] It can be understood that in the above step S123c, the specific method for determining the cell state vector corresponding to the target photovoltaic cell (i.e., fusing three image semantics) is not limited and can be selected according to actual needs. For example, in an alternative embodiment, in order to more reliably fuse the three image semantics and make the determined cell state vector more reliable, the above step S123c can further include steps c1, c2, c3, c4, and c5, and the specific content of each step is as follows.

[0093] Step c1: Perform an association mining operation on each local vector in the cell state vector corresponding to the image member set according to the sequence relationship corresponding to each image member in the image member set, and output the first association mining vector corresponding to the image member set.

[0094] In an embodiment of the present invention, an association mining operation can be performed on each local vector in the cell state vector corresponding to the image member set according to the sequence relationship corresponding to each image member in the image member set, and the first association mining vector corresponding to the image member set can be output. Exemplarily, in combination with Figure 4, for each image member other than the first image member in the set of image members (such as the second one), the output vector corresponding to the previous image member (such as the first one) can be multiplied by the local vector corresponding to the current image member (such as the second one) to obtain the corresponding correlation characterization parameter. Then, based on this correlation characterization parameter, the local vector is weighted to obtain the corresponding weighted vector. Also, the weighted vector and the local vector are subjected to a superposition operation to obtain the corresponding output vector. Among them, the output vector corresponding to the first image member is the corresponding local vector, and the output vector corresponding to the last image member is used as the first correlation mining vector corresponding to the set of image members.

[0095] Step c2: According to the sequence relationship corresponding to each image member in the set of image members, perform a difference operation on adjacent local vectors in the cell state vector corresponding to the set of image members, and output the corresponding set of local difference vectors.

[0096] In the embodiment of the present invention, according to the sequence relationship corresponding to each image member in the set of image members, a difference operation on adjacent local vectors in the cell state vector corresponding to the set of image members can be performed to output the corresponding set of local difference vectors. Exemplarily, the second local vector can be subtracted from the first local vector to obtain the first local difference vector, the third local vector can be subtracted from the second local vector to obtain the second local difference vector, the fourth local vector can be subtracted from the third local vector to obtain the third local difference vector, and so on. In this way, the local difference vectors in the set of local difference vectors can carry the change information between the historical cell images, and this change information has a certain characterization ability for the defect states that appear. In addition, in other embodiments, combined with Figure 5 , it is also possible to perform image difference calculation on the cell images corresponding to each image member in the set of image members (that is, calculate the difference between the pixel values at the corresponding positions of the subsequent frame and the previous frame). In this way, a set of difference images can be obtained. Then, for each frame of difference image in the set of difference images, perform the feature space mapping operation and deep semantic mining operation as described above to obtain the corresponding set of local difference vectors.

[0097] Step c3: According to the sequence relationship corresponding to each image member in the set of image members, perform a correlation mining operation on each local difference vector in the set of local difference vectors, and output the second correlation mining vector corresponding to the set of image members.

[0098] In an embodiment of the present invention, the local difference vector set can be subjected to an association mining operation according to the sequence relationship corresponding to each image member in the image member set, and a second association mining vector corresponding to the image member set is output. The specific manner of the association mining operation is as described above.

[0099] Step c4: Aggregate the first association mining vector corresponding to the image member set and the second association mining vector corresponding to the image member set through a vector aggregation operation, and output an aggregated association mining vector corresponding to the image member set.

[0100] In an embodiment of the present invention, the first association mining vector corresponding to the image member set and the second association mining vector corresponding to the image member set can be subjected to a vector aggregation operation to output an aggregated association mining vector corresponding to the image member set. Exemplarily, the first association mining vector and the second association mining vector can be processed by superposition, averaging, splicing, etc. to achieve corresponding aggregation, thereby obtaining the corresponding aggregated association mining vector.

[0101] Step c5: Based on the aggregated association mining vector corresponding to the image member set and the cell state vectors corresponding to each of the at least one related cell member, perform an association focusing mining operation on the cell state vector corresponding to the target cell member, and output the cell state vector corresponding to the target photovoltaic cell.

[0102] In an embodiment of the present invention, the cell state vector corresponding to the target cell member can be subjected to an association focusing mining operation based on the aggregated association mining vector corresponding to the image member set and the cell state vectors corresponding to each of the at least one related cell member, and the cell state vector corresponding to the target photovoltaic cell is output. That is, it is necessary to fuse the semantic information represented by the aggregated association mining vector corresponding to the image member set and the cell state vectors corresponding to each of the at least one related cell member into the cell state vector corresponding to the target cell member, so that the semantic information of the cell state vector corresponding to the target photovoltaic cell is richer.

[0103] It can be understood that in step c5 above, the specific manner of performing the association focusing mining operation is not limited and can be selected according to actual needs. For example, in an alternative embodiment, in order to enable the semantic information represented by multiple vectors to be fully fused, so as to obtain a more reliable cell state vector, combined with Figure 6 , step c5 above may include:

[0104] First, the aggregation correlation mining vector corresponding to the set of image members and the cell state vectors corresponding to the at least one relevant cell member can be respectively used as vectors to be aggregated, so as to obtain a plurality of vectors to be aggregated. That is to say, the plurality of vectors to be aggregated are respectively the aggregation correlation mining vector corresponding to the set of image members and the cell state vectors corresponding to the at least one relevant cell member as vectors to be aggregated;

[0105] Second, for each of the vectors to be aggregated, an association analysis operation can be performed on the vector to be aggregated and the cell state vector corresponding to the target cell member, and a correlation characterization parameter corresponding to the vector to be aggregated can be output. Based on the correlation characterization parameter, an update operation is performed on the vector to be aggregated, and an associated focused mining vector corresponding to the vector to be aggregated is output. Exemplarily, the vector to be aggregated and the cell state vector can be multiplied to obtain a corresponding correlation characterization parameter. Then, based on the correlation characterization parameter, the vector to be aggregated is weighted, so as to obtain a corresponding associated aggregation mining vector. In this way, the semantic information represented by the associated aggregation mining vector can be more focused on the features related to the cell state vector, that is, the cell state vector is fused into the vector to be aggregated;

[0106] Then, the central vector of the plurality of associated focused mining vectors corresponding to the plurality of vectors to be aggregated can be determined, and the central vector and the cell state vector corresponding to the target cell member are subjected to a superposition operation to output the cell state vector corresponding to the target photovoltaic cell. Exemplarily, the mean vector of the plurality of associated focused mining vectors can be calculated as the central vector; the mean vector of the plurality of associated focused mining vectors can also be calculated, and then the associated focused mining vector with the smallest cosine distance from the mean vector among the plurality of associated focused mining vectors is used as the central vector. It is also possible to perform clustering processing on the plurality of associated focused mining vectors to obtain corresponding multiple cluster centers (the specific number can be configured according to actual needs, such as numerical values of 2, 3, 4, etc.), and then calculate the mean of the multiple cluster centers to obtain the central vector.

[0107] Regarding step S130, it should be noted that the specific method for analyzing at least one photovoltaic cell state information corresponding to the target photovoltaic cell is not limited either and can be selected according to actual needs. For example, in an alternative implementation manner, in order to reliably analyze at least one photovoltaic cell state information corresponding to the target photovoltaic cell, step S130 described above can further include step S131, step S132, and step S133, and the specific content of each step is as follows.

[0108] Step S131: Obtain a cell state vector corresponding to each of at least one photovoltaic cell state information to be confirmed.

[0109] In an embodiment of the present invention, a cell state vector corresponding to each of at least one photovoltaic cell state information to be confirmed may be obtained first, where the cell state vector is used to reflect the semantic information of the corresponding photovoltaic cell state information to be confirmed.

[0110] Step S132: For each of the at least one photovoltaic cell state information to be confirmed, determine a vector association parameter between the cell state vector corresponding to the target photovoltaic cell and the cell state vector corresponding to the photovoltaic cell state information.

[0111] In an embodiment of the present invention, for each of the at least one photovoltaic cell state information to be confirmed, a vector association parameter between the cell state vector corresponding to the target photovoltaic cell and the cell state vector corresponding to the photovoltaic cell state information is determined. Exemplarily, a cosine similarity between the cell state vector corresponding to the target photovoltaic cell and the cell state vector corresponding to the photovoltaic cell state information may be calculated as the corresponding vector association parameter.

[0112] Step S133: Among the at least one photovoltaic cell state information to be confirmed, determine the photovoltaic cell state information whose vector association parameter is greater than a pre-determined reference vector association parameter, and determine this photovoltaic cell state information as the photovoltaic cell state information corresponding to the target photovoltaic cell.

[0113] In an embodiment of the present invention, among the at least one photovoltaic cell state information to be confirmed, the photovoltaic cell state information whose vector association parameter is greater than a pre-determined reference vector association parameter may be determined, and this photovoltaic cell state information may be determined as the photovoltaic cell state information corresponding to the target photovoltaic cell. In this way, at least one photovoltaic cell state information may be obtained, such as having defects such as warping and cracking simultaneously.

[0114] It can be understood that in the above step S131, the specific manner of obtaining a cell state vector corresponding to each of at least one photovoltaic cell state information to be confirmed is not limited and may be selected according to actual needs. For example, in an alternative embodiment, in order to make the semantic information of the cell state vectors corresponding to the photovoltaic cell state information richer and more accurate, the above step S131 may further include:

[0115] First, for each photovoltaic cell status information among at least one photovoltaic cell status information to be confirmed, at least one photovoltaic cell status associated information corresponding to the photovoltaic cell status information can be extended, where there is a related defect status semantics between the photovoltaic cell status associated information and the photovoltaic cell status information, such as synonyms or synonymous sentences, etc.;

[0116] Second, the first feature space projection unit included in the status information mining model can be used to perform a feature space projection operation on at least one status information unit included in the photovoltaic cell status information, and output a status information projection vector corresponding to the photovoltaic cell status information. Exemplarily, the first feature space projection unit can be a word embedding model, and the status information unit can refer to the words obtained by performing word segmentation on the photovoltaic cell status information, such as "has", "relatively wide", "of", "crack". In this way, word embedding processing can be performed on each word respectively to obtain corresponding word embedding vectors, and then the word embedding vectors can be combined, such as concatenation, to form a corresponding status information projection vector. For example, the word embedding vectors embeddings = {"has": [0.12, 0.45, -0.78, 0.36,...], "relatively wide": [0.32, 0.54, -0.34, 0.29,...], "of": [0.15, -0.04, 0.22, 0.51,...], "crack": [-0.41, 0.22, 0.87, -0.67,...]};

[0117] Then, the second feature space projection unit included in the status information mining model can be used to perform a feature space projection operation on at least one status information unit included in each of the at least one photovoltaic cell status associated information respectively, and output a status information projection vector corresponding to the photovoltaic cell status associated information; Exemplarily, the second feature space projection unit can be the same word embedding model as the first feature space projection unit;

[0118] Furthermore, the vector aggregation unit included in the status information mining model can be used to perform a vector aggregation operation on the status information projection vector corresponding to the photovoltaic cell status information and the status information projection vector corresponding to the photovoltaic cell status associated information, and output a corresponding status information aggregation vector; Exemplarily, each status information projection vector can be processed by superposition or concatenation, etc., to achieve vector aggregation and obtain a corresponding status information aggregation vector;

[0119] Finally, the deep semantic representation unit included in the state information mining model can be used to perform deep semantic representation operations on the state information aggregation vector, and output the cell state vector corresponding to the photovoltaic cell state information; exemplarily, the deep semantic representation operation may refer to self-attention processing, and correspondingly, the deep semantic representation unit may be a self-attention network.

[0120] In addition, it should be noted that the state information mining model can be trained together with the aforementioned photovoltaic cell mining model. That is to say, first, a sample device relationship graph can be determined. Then, based on the sample device relationship graph, the cell state vector corresponding to the corresponding sample photovoltaic cell is mined through the initial photovoltaic cell mining model, and at least one cell state vector corresponding to the photovoltaic cell state information to be confirmed is mined through the initial state information mining model. Then, at least one photovoltaic cell state information corresponding to the sample photovoltaic cell is determined. Then, an error calculation is performed between the at least one photovoltaic cell state information and the actual state (defect label, which can be formed by manual annotation) of the sample photovoltaic cell. Then, along the direction of reducing the calculated error, the model parameters of the initial state information mining model and the initial photovoltaic cell mining model are adjusted until the error converges, so as to obtain the final state information mining model and photovoltaic cell mining model, so that the above steps S120 and S130 can be executed through the final state information mining model and photovoltaic cell mining model, thereby realizing reliable detection of the defect state of the photovoltaic device.

[0121] In addition, for the above cell image, it can be a segmented image. That is to say, after the original image is acquired through an image acquisition device such as an industrial camera, the original image may include multiple photovoltaic cells. Therefore, the original image can be segmented through an image segmentation model (the specific composition of the image segmentation model can refer to relevant existing technologies and will not be specifically limited here), so as to form the cell image corresponding to each photovoltaic cell. Or, in some application scenarios, when the granularity or size of the photovoltaic cell is large, the acquired original image can also be directly used as the cell image corresponding to the photovoltaic cell.

[0122] In addition, for the above series connection or parallel connection, it can be determined either by manual annotation or by the distribution positions between the corresponding photovoltaic cells. For example, the photovoltaic cells in the same column can be connected in series, and the photovoltaic cells in different columns can be connected in parallel. These can be configured accordingly according to the actual situation.

[0123] In summary, the photovoltaic device detection method and system provided by the present invention can determine an equipment relationship map based on at least one cell image corresponding to a photovoltaic cell; secondly, determine a cell state vector corresponding to the target photovoltaic cell based on the cell relationship map corresponding to the target photovoltaic cell in the equipment relationship map; then, analyze at least one photovoltaic cell state information corresponding to the target photovoltaic cell based on the matching relationship between the cell state vector corresponding to the target photovoltaic cell and the cell state vectors corresponding to at least one photovoltaic cell state information to be confirmed. Based on the above content, since the cell relationship map corresponding to the target photovoltaic cell includes the target cell members corresponding to the target photovoltaic cell and at least one image member connected by the relationship representation line by the target cell members, when determining the cell state vector based on the cell relationship map, not only the information of the target cell members is referred to, but also the information of the image members is referred to, so that the semantic information included in the cell state vector is richer. Therefore, when analyzing the photovoltaic cell state information based on the cell state vector, a more comprehensive analysis basis can be obtained, making the reliability of the analysis result higher, that is, the reliability of the obtained photovoltaic cell state information is higher. Therefore, the problem of relatively low reliability in photovoltaic device detection existing in the prior art can be improved.

[0124] In several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device and method embodiments described above are only illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of devices, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, program segment, or part of code includes one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0125] In addition, the various functional modules in the embodiments of the present invention can be integrated together to form an independent part, or each module can exist alone, or two or more modules can be integrated to form an independent part.

[0126] When the above-mentioned functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs. It should be noted that in this article, the terms "including", "comprising", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article, or device. Without further limitations, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article, or device including the said element.

[0127] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A photovoltaic equipment detection method based on machine vision, characterized in that: include: Determine a device relationship map based on a cell image corresponding to at least one photovoltaic cell, wherein the device relationship map includes a cell member corresponding to the photovoltaic cell, an image member corresponding to the cell image, and a relationship representation line, wherein the relationship representation line is used to reflect that there is a correlation between the photovoltaic cell corresponding to the cell member and the cell image corresponding to the image member; Based on the cell relationship graph corresponding to the target photovoltaic cell in the device relationship graph, a cell state vector corresponding to the target photovoltaic cell is determined, wherein the cell relationship graph corresponding to the target photovoltaic cell includes a target cell member corresponding to the target photovoltaic cell and at least one image member connected to the target cell member by a relationship characterization line, the cell state vector is used to reflect the state semantics of the target photovoltaic cell, the cell image corresponding to the image member refers to a historical cell image formed by historically performing image acquisition on the photovoltaic cell corresponding to the target cell member, the historical cell image is used as member attribute data of the image member in the device relationship graph, the member attribute data of the target cell member in the device relationship graph includes a current cell image formed by currently performing image acquisition on the corresponding photovoltaic cell, and the cell state vector carries change information between each cell image in history; Based on the matching relationship between the cell state vector corresponding to the target photovoltaic cell and the cell state vector corresponding to at least one photovoltaic cell state information to be confirmed, at least one photovoltaic cell state information corresponding to the target photovoltaic cell is analyzed, wherein the cell state vector corresponding to the photovoltaic cell state information is used to reflect the defect state semantics of the photovoltaic cell state information.

2. The photovoltaic equipment detection method based on machine vision according to claim 1, characterized in that: The step of determining the cell state vector corresponding to the target photovoltaic cell based on the cell relationship graph corresponding to the target photovoltaic cell in the device relationship graph includes: Determining an image member set based on at least one image member in the cell relationship graph connected to the target cell member through the relationship representation line; Determine a cell state vector corresponding to the image member set, wherein the cell state vector corresponding to the image member set is used to reflect the state semantics of at least one cell image having a correlation relationship with the target photovoltaic cell; Based on the cell state vector corresponding to the image member set, the cell state vector corresponding to the target photovoltaic cell is determined.

3. The photovoltaic equipment detection method based on machine vision according to claim 2, characterized in that: The step of determining the image member set based on at least one image member having the relationship characterization line with the target cell member in the cell relationship map comprises: For each image member of at least one image member connected to the target cell member through a relationship characterization line, based on the image acquisition time configured on the relationship characterization line between the image member and the target cell member, a sequential relationship of the image member in the image member set is obtained; According to the corresponding sequential relationship, at least one image member connected to the target cell member through the relationship characterization line is combined to form a corresponding image member set.

4. The photovoltaic equipment detection method based on machine vision according to claim 2, characterized in that: The step of determining the cell state vector corresponding to the image member set includes: Acquire member attribute data corresponding to at least one image member included in the image member set, wherein the member attribute data corresponding to the image member at least includes a cell image corresponding to the image member; Loading member attribute data corresponding to each of the at least one image member so that the photovoltaic cell mining model acquires the member attribute data; Using the feature space mapping unit included in the photovoltaic cell mining model, the member attribute data corresponding to each of the at least one image member is subjected to a feature space mapping operation, and a photovoltaic image mapping vector corresponding to the image member set is output; The deep semantic mining unit in the photovoltaic cell mining model is used to perform a deep semantic mining operation on the photovoltaic image mapping vector corresponding to the image member set, and the cell state vector corresponding to the image member set is output.

5. The photovoltaic equipment detection method based on machine vision according to claim 2, characterized in that: The cell relationship map also includes at least one related cell member connected to the target cell member through a relationship characterization line, and the photovoltaic cell corresponding to the related cell member has a correlation relationship in electrical properties with the photovoltaic cell corresponding to the target cell member, and the correlation relationship includes series connection and / or parallel connection; The step of determining the cell state vector corresponding to the target photovoltaic cell based on the cell state vector corresponding to the image member set includes: Using the feature space mapping unit included in the photovoltaic cell mining model, feature space mapping operations are performed on the member attribute data corresponding to the target cell member corresponding to the target photovoltaic cell and the member attribute data corresponding to at least one related cell member connected to the target cell member through a relationship characterization line, and a photovoltaic image mapping vector corresponding to the target cell member and a photovoltaic image mapping vector corresponding to the at least one related cell member are output; Utilizing the deep semantic mining unit in the photovoltaic cell mining model, respectively performing deep semantic mining operations on the photovoltaic image mapping vector corresponding to the target cell member and the photovoltaic image mapping vector corresponding to each of the at least one related cell member, and outputting the cell state vector corresponding to the target cell member and the cell state vector corresponding to each of the at least one related cell member; The cell state vector corresponding to the target photovoltaic cell is determined based on the cell state vector corresponding to the image member set, the cell state vector corresponding to the target cell member and the cell state vector corresponding to each of the at least one related cell member.

6. The photovoltaic equipment detection method based on machine vision according to claim 5, characterized in that: The step of determining the cell state vector corresponding to the target photovoltaic cell based on the cell state vector corresponding to the image member set, the cell state vector corresponding to the target cell member, and the cell state vector corresponding to each of the at least one related cell members comprises: According to the order relationship between the image members in the image member set, an association mining operation is performed on each local vector in the cell state vector corresponding to the image member set, and a first association mining vector corresponding to the image member set is output; According to the order relationship of the image members in the image member set, each local vector in the cell state vector corresponding to the image member set is subjected to adjacent difference operation, and a corresponding local difference vector set is output; According to the order relationship of the image members in the image member set, an association mining operation is performed on each local difference vector in the local difference vector set, and a second association mining vector corresponding to the image member set is output; Performing a vector aggregation operation on a first association mining vector corresponding to the image member set and a second association mining vector corresponding to the image member set, and outputting an aggregated association mining vector corresponding to the image member set; Based on the aggregated association mining vector corresponding to the image member set and the cell state vector corresponding to each of the at least one related cell members, an association focusing mining operation is performed on the cell state vector corresponding to the target cell member, and the cell state vector corresponding to the target photovoltaic cell is output.

7. The photovoltaic equipment detection method based on machine vision according to claim 6, characterized in that: The step of performing an association focused mining operation on the cell state vector corresponding to the target cell member based on the aggregated association mining vector corresponding to the image member set and the cell state vector corresponding to each of the at least one related cell members, and outputting the cell state vector corresponding to the target photovoltaic cell, comprises: Respectively taking the aggregated association mining vector corresponding to the image member set and the cell state vector corresponding to each of the at least one related cell members as vectors to be aggregated, so as to obtain a plurality of vectors to be aggregated; For each of the vectors to be aggregated, an association analysis operation is performed between the vector to be aggregated and the cell state vector corresponding to the target cell member, and a correlation characterization parameter corresponding to the vector to be aggregated is output; based on the correlation characterization parameter, an update operation is performed on the vector to be aggregated, and an association focus mining vector corresponding to the vector to be aggregated is output; The central vector of the multiple associated focused mining vectors corresponding to the multiple vectors to be aggregated is determined, and the central vector and the cell state vector corresponding to the target cell member are superimposed to output the cell state vector corresponding to the target photovoltaic cell.

8. The photovoltaic equipment detection method based on machine vision according to any one of claims 1 to 7, characterized in that: The step of analyzing at least one photovoltaic cell state information corresponding to the target photovoltaic cell based on the matching relationship between the cell state vector corresponding to the target photovoltaic cell and the cell state vector corresponding to at least one photovoltaic cell state information to be confirmed comprises: Obtaining a cell state vector corresponding to at least one photovoltaic cell state information to be confirmed; For each photovoltaic cell state information in at least one photovoltaic cell state information to be confirmed, determining a vector association parameter between a cell state vector corresponding to the target photovoltaic cell and a cell state vector corresponding to the photovoltaic cell state information; Among the at least one photovoltaic cell state information to be confirmed, determine the photovoltaic cell state information to be confirmed whose vector associated parameter is greater than a predetermined reference vector associated parameter, and determine the photovoltaic cell state information as the photovoltaic cell state information corresponding to the target photovoltaic cell.

9. The photovoltaic equipment detection method based on machine vision according to claim 8, characterized in that: The step of obtaining a cell state vector corresponding to at least one photovoltaic cell state information to be confirmed comprises: For each photovoltaic cell state information in at least one photovoltaic cell state information to be confirmed, at least one photovoltaic cell state association information corresponding to the photovoltaic cell state information is expanded, wherein the photovoltaic cell state association information and the photovoltaic cell state information have related defect state semantics; Using the first feature space projection unit included in the state information mining model, at least one state information unit included in the photovoltaic cell state information is subjected to a feature space projection operation, and a state information projection vector corresponding to the photovoltaic cell state information is output; Using the second feature space projection unit included in the state information mining model, respectively perform a feature space projection operation on at least one state information unit included in each of the at least one photovoltaic cell state association information, and output a state information projection vector corresponding to the photovoltaic cell state association information; Using the vector aggregation unit included in the state information mining model, the state information projection vector corresponding to the photovoltaic cell state information and the state information projection vector corresponding to the photovoltaic cell state association information are subjected to vector aggregation operation, and the corresponding state information aggregation vector is output; The deep semantic representation unit included in the state information mining model is used to perform a deep semantic representation operation on the state information aggregation vector, and a cell state vector corresponding to the photovoltaic cell state information is output.

10. A photovoltaic equipment detection system based on machine vision, characterized in that: It comprises a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the photovoltaic device detection method based on machine vision as described in any one of claims 1 to 9.

Images