Photovoltaic module crack detection method and related products

Abstract

The embodiment of the application discloses the field of artificial intelligence, and particularly relates to a photovoltaic module hidden crack detection method and related products in the field of photovoltaic module defect detection, and the method comprises: hidden crack feature extraction processing is carried out on a first image to obtain a second image; the hidden crack feature extraction processing comprises: adjusting the gray value of a first pixel point to a first gray value, and adjusting the gray value of a second pixel point to a second gray value; a first eigenvalue of a first Hessian matrix corresponding to the first pixel point satisfies a target condition, a second eigenvalue of a second Hessian matrix corresponding to the second pixel point does not satisfy the target condition, and the first gray value is obtained by using the first eigenvalue; a defect recognition model trained is used for defect recognition on the second image, and a defect recognition result is output; the defect recognition result indicates that the photovoltaic module has hidden cracks or does not have hidden cracks; the accuracy of recognizing hidden crack defects in the photovoltaic module can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of photovoltaic module defect detection, and in particular to a photovoltaic module hidden crack detection method and related products. BACKGROUND

[0002] A photovoltaic system is a system that uses photovoltaic modules (also known as solar cell modules) and other auxiliary equipment to convert solar energy into electrical energy. At present, photovoltaic systems are widely used and have played a great role in promoting market expansion. In particular, after the adjustment in 2018, the global photovoltaic system market has returned to a strong development track.

[0003] As a core component of a photovoltaic system, a photovoltaic module plays a crucial role in the power generation efficiency of the entire system. However, in addition to the unevenness of the manufacturing process, on-site construction and installation, and post-operation maintenance can also be factors that affect the quality of the photovoltaic module. Therefore, regular detection of the photovoltaic module is an essential procedure, such as electro luminescence (EL) detection of the photovoltaic module.

[0004] EL detection of a photovoltaic module is to apply a reverse voltage to the photovoltaic module and take a photo of it in a dark environment using a professional camera. Internal defects such as hidden cracks, broken grids, broken screens, and black spots can be found from the photographed image. Hidden cracks are one of the most common defects of photovoltaic modules. On the one hand, they can affect the power generation of the photovoltaic system. On the other hand, hidden cracks can gradually worsen during operation and become more serious defects, and in the worst case, they can pose a threat to the safety of the entire photovoltaic system. Therefore, a solution that can accurately identify hidden crack defects in photovoltaic modules needs to be researched. SUMMARY

[0005] The embodiments of the present application disclose a photovoltaic module hidden crack detection method and related products, which can improve the accuracy of identifying hidden crack defects in photovoltaic modules.

[0006] In a first aspect, an embodiment of the present application provides a method for detecting a hidden crack in a photovoltaic module. The method comprises: performing a hidden crack feature extraction process on a first image to obtain a second image; the first image is an electroluminescence (EL) image of the photovoltaic module obtained by detection; the hidden crack feature extraction process comprises: adjusting a gray value of a first pixel point to a first gray value, and adjusting a gray value of a second pixel point to a second gray value; the first pixel point and the second pixel point are both included in the first image; a first eigenvalue of a first Hessian matrix corresponding to the first pixel point satisfies a target condition, a second eigenvalue of a second Hessian matrix corresponding to the second pixel point does not satisfy the target condition; the first gray value is obtained by using the first eigenvalue; the second gray value is different from the first gray value; performing defect recognition on the second image by using a trained defect recognition model, and outputting a defect recognition result; the defect recognition result indicates whether the photovoltaic module has a hidden crack or not.

[0007] An execution subject of an embodiment of the present application is a hidden crack detection device. The first image can be an EL image of the photovoltaic module (for example, a photovoltaic cell) obtained by using an infrared camera to take a picture in a dark room after an external voltage is applied to the photovoltaic module. In an embodiment of the present application, the first image can be obtained in various ways, which are not limited in the present application. In a possible implementation, the second gray value is 0.

[0008] The grains in the EL image of the photovoltaic module are usually in a spot shape, and the hidden crack has a strong linear feature. Research shows that there is a great difference between the eigenvalues of the Hessian matrices corresponding to the pixel points (for example, the first pixel point and the second pixel point) corresponding to the two different structures (that is, the grains and the hidden crack). The first pixel point can be understood as any pixel point corresponding to the hidden crack defect, and the second pixel point can be understood as any pixel point not corresponding to the hidden crack defect. The eigenvalue of the Hessian matrix corresponding to the pixel point corresponding to the hidden crack defect in the first image usually satisfies the target condition, and the eigenvalue of the Hessian matrix corresponding to the pixel point not corresponding to the hidden crack defect usually does not satisfy the target condition. In an embodiment of the present application, the gray value of the pixel point is adjusted according to whether the eigenvalue of the Hessian matrix corresponding to the pixel point in the first image satisfies the target condition, so that the hidden crack feature in the first image can be accurately extracted.

[0009] In an embodiment of the present application, the gray value of the first pixel point is adjusted to the first gray value, and the gray value of the second pixel point is adjusted to the second gray value, so that the hidden crack feature in the first image can be accurately extracted, thereby improving the accuracy of recognizing the hidden crack defect in the photovoltaic module.

[0010] In a possible implementation, the first characteristic value satisfying the target condition includes that a third characteristic value included in the first characteristic value is less than or equal to 0, the third characteristic value being a characteristic value with a larger absolute value in two characteristic values included in the first characteristic value, and the second characteristic value not satisfying the target condition includes that a fourth characteristic value included in the second characteristic value is greater than 0, the fourth characteristic value being a characteristic value with a larger absolute value in two characteristic values included in the second characteristic value.

[0011] The first characteristic value can include each characteristic value of the first Hessian matrix, and the second characteristic value can include each characteristic value of the second Hessian matrix. The first Hessian matrix and the second Hessian matrix can each be a 2-row 2-column matrix. For a set of characteristic values (for example, the first characteristic value) of a Hessian matrix corresponding to any pixel point in the first image, if a characteristic value with a larger absolute value in the set of characteristic values is less than or equal to 0, the set of characteristic values satisfies the target condition; otherwise, the set of characteristic values does not satisfy the target condition. The first pixel point can be understood as any pixel point corresponding to a hidden crack defect, and the second pixel point can be understood as any pixel point not corresponding to a hidden crack defect. It should be understood that whether a pixel point corresponds to a hidden crack defect can be determined according to whether the characteristic value of the Hessian matrix corresponding to the pixel point satisfies the target condition, so that the hidden crack feature can be accurately extracted.

[0012] In this implementation, when a third characteristic value included in the first characteristic value is less than or equal to 0, the first characteristic value satisfies the target condition; when a fourth characteristic value included in the second characteristic value is greater than 0, the second characteristic value does not satisfy the target condition; and the pixel point corresponding to the hidden crack defect and the pixel point not corresponding to the hidden crack defect can be accurately distinguished.

[0013] In a possible implementation, before the gray value of the first pixel point is adjusted to the first gray value and the gray value of the second pixel point is adjusted to the second gray value, the method further includes: constructing the first Hessian matrix corresponding to the first pixel point and constructing the second Hessian matrix corresponding to the second pixel point; processing the first Hessian matrix to obtain the first characteristic value, and processing the second Hessian matrix to obtain the second characteristic value.

[0014] Hereinafter, the first Hessian matrix of the first pixel point is taken as an example to introduce the Hessian matrix corresponding to any pixel point in the first image.

[0015] For example, the first Hessian matrix of the first pixel point is as follows:

[0016]

[0017] wherein, I xx represents the second-order partial derivative of the first pixel point in the x direction, I yydenotes the second-order partial derivative of the first pixel point in the x direction, I xy = I yx is the mixed partial derivative of the first pixel point in the xy direction. Specifically as follows:

[0018] I xx = I(i+1, j) - 2*I(i, j) + I(i-1, j) (2);

[0019] I yy = I(i, j+1) - 2*I(i, j) + I(i, j-1) (3);

[0020] I xy = I(i+1, j+1) - I(i, j+1) - I(i+1, j) + I(i, j) (4);

[0021] wherein I(i, j) is the gray value of the first pixel point, i.e. the gray value of the pixel point with coordinates (i, j) in the first image, I(i, j-1) is the gray value of the pixel point with coordinates (i, j-1) in the first image, and so on. In actual application, the hidden crack detection device can calculate the Hessian matrix corresponding to each pixel point in the first image by using similar formulas (2) to (4), or by using other methods.

[0022] In this implementation, the first Hessian matrix corresponding to the first pixel point is constructed, and the second Hessian matrix corresponding to the second pixel point is constructed, and then the eigenvalues required for hidden crack feature extraction processing of the first pixel point and the second pixel point are obtained, so as to accurately distinguish the pixel points corresponding to hidden crack defects and the pixel points not corresponding to hidden crack defects.

[0023] In one possible implementation, the following formula is used for hidden crack feature extraction processing of the first image:

[0024]

[0025] wherein R b = λ1 / λ2, S = λ1 2 + λ2 2 , λ1, λ2 are eigenvalues (e.g. first eigenvalue or second eigenvalue) of the Hessian matrix corresponding to any pixel point u in the first image, β and c are fixed parameters for controlling sensitivity, is the gray value of any pixel point u in the first image after hidden crack feature extraction. β and c can be configured according to actual needs.

[0026] In this implementation, the formula (5) can be used to accurately perform the hidden crack feature extraction on the first image, that is, to accurately distinguish the pixel points corresponding to the hidden crack defects and the pixel points not corresponding to the hidden crack defects.

[0027] In one possible implementation, after the hidden crack feature extraction is performed on the first image to obtain the second image, the method further includes: performing a hidden crack feature enhancement on the second image to obtain a third image; the hidden crack feature enhancement includes: obtaining F gray scale responses of a third pixel point in the second image in K directions, and adjusting a gray scale value of the third pixel point to a third gray scale value; the third gray scale value is obtained by using the F gray scale responses, and a gray scale response of the third pixel point in a first direction is used to determine whether there is a hidden crack in the first direction of the third pixel point, the first direction is included in the K directions, K is an integer greater than 1, and F is an integer equal to or greater than K; the defect recognition on the second image by using the trained defect recognition model to output the defect recognition result includes: performing the defect recognition on the third image by using the trained defect recognition model to output the defect recognition result. Since the hidden crack in the image usually has a linear feature, the F gray scale responses of the third pixel point in the K directions can further enhance the hidden crack feature.

[0028] In this implementation, adjusting the gray scale value of the third pixel point to the third gray scale value can enhance the hidden crack feature in the image.

[0029] In a possible implementation, the obtaining the F gray scale responses of the third pixel point in the second image in the K directions includes: obtaining a first gray scale matrix; the first gray scale matrix is obtained by point multiplication of a sliding matrix and a first sub-matrix, a first element in the sliding matrix is not zero, a second element in the sliding matrix is zero, a direction of a first ray corresponding to the first element is included in the K directions, the first ray is a ray in the sliding matrix starting from a center element and passing through the first element, a direction of a second ray corresponding to the second element is not included in the K directions, the second ray is a ray in the sliding matrix starting from the center element and passing through the second element, a center element of the first sub-matrix is the third pixel point, and the first sub-matrix is a sub-matrix of the second image; the F gray scale responses of the third pixel point in the K directions are obtained according to the first gray scale matrix; K rays starting from a center element in the first gray scale matrix correspond to the K directions one by one, the first direction corresponds to a third ray in the K rays, the third ray passes through two or more elements in the first gray scale matrix, and the gray scale response of the third pixel point in the first direction includes the third ray passing through the two or more elements in the first gray scale matrix.

[0030] In this implementation, the F gray scale responses of the third pixel point in the K directions can be accurately obtained according to the first gray scale matrix.

[0031] In a possible implementation, the directions of the K rays starting from the center element in the sliding matrix correspond to the K directions one by one, and an included angle between any two adjacent rays in the K rays starting from the center element in the sliding matrix is a target angle.

[0032] For example, K is 12, and the target angle is 24. For another example, K is 18, and the target angle is 20.

[0033] In this implementation, the included angle between any two adjacent rays in the K rays is the target angle, so that the gray scale responses in each direction are detected.

[0034] In a possible implementation, the first gray scale matrix is an M-row and M-column matrix, the third ray passes through L elements in the first gray scale matrix, M is an odd number greater than 1, and L is an integer greater than 0 and not greater than M; after the F gray scale responses of the third pixel point in the K directions are obtained according to the first gray scale matrix, the method further includes: processing the F gray scale responses of the third pixel point in the K directions to obtain the third gray scale value. F=L*K. The third pixel point corresponds to L gray scale responses in each of the K directions.

[0035] In some embodiments, the F gray scale responses of the third pixel point in the K directions and the third gray scale value satisfy the following formula:

[0036]

[0037] wherein, represents the third gray scale value, represents the maximum gray scale response of the third pixel point in the K directions, represents the average value of the F gray scale responses, and d is a constant, for example, d is 1. The value of d can be configured as needed. may be the maximum value of the K first sub-gray scale responses in the K directions. The K first sub-gray scale responses correspond to the K directions one by one, and the first sub-gray scale response in each direction is obtained by using the L gray scale responses in the direction. For example, the first sub-gray scale response in a direction is the weighted sum of the F gray scale responses in the direction. For another example, the first sub-gray scale response in a direction is the maximum value of the F gray scale responses in the direction. The L gray scale responses in any direction of the K directions are the L elements in the first gray scale matrix passed by the ray corresponding to the direction. For example, the L elements in the first gray scale matrix passed by the third ray in turn are a, b, c, d, e, and the L gray scale responses in the direction corresponding to the third ray are a, b, c, d, e; the (L-1) gray scale responses in the direction corresponding to the third ray are a, b, c, d, and the (L-2) gray scale responses in the direction corresponding to the third ray are a, b, c; wherein L is 5. may be obtained by using the L*K gray scale responses (i.e., the L gray scale responses in each direction) of the third pixel point in the K directions. may be understood as the gray scale response of the linear detector with a length of L to the third pixel point. The hidden crack detection device can also obtain the gray scale response of the linear detector with a length of (L-1) to the third pixel point by using the (L-1)*K gray scale responses (i.e., the (L-1) gray scale responses in each direction) of the third pixel point in the K directions.

[0038] The (L-1)*K gray scale responses of the third pixel point in the K directions can satisfy the following formula:

[0039]

[0040] wherein, represents the gray scale response of the linear detector with a length of (L-1) to the third pixel point, represents a maximum value in K second sub-gray responses in the K directions, represents an average value of the (L-1)*K gray responses, d is a constant, for example, d is 1. The K second sub-gray responses correspond to the K directions one by one, and the second sub-gray response in each direction is obtained by using (L-1) gray responses in the direction. It should be understood that the crack detection device can obtain the gray responses of two or more linear detectors with different lengths for the third pixel point by processing the first gray matrix, for example and

[0041] To increase the robustness of the linear detector, different lengths of linear detectors can be selected in the first gray matrix for calculation, and finally the gray responses of each linear detector for the third pixel point obtained by calculation are weighted to obtain the third gray value. In some embodiments, the F gray responses of the third pixel point in the K directions and the third gray value satisfy the following formula:

[0042]

[0043] wherein R represents the third gray value, represents the gray response of the linear detector with length L for the third pixel point, P is the gray value of the third pixel point, and n is the scale size of the linear detector (i.e., the number of linear detectors). represents the sum.

[0044] It should be understood that after the crack detection device obtains the gray responses of any pixel point in the K directions, it can calculate the gray value of other pixel points in the second image to be adjusted to in a similar manner to the calculation of the third gray value. Since the hidden crack defect has directionality within a certain range, by selecting linear detectors in different directions within a fixed sliding window to calculate the gray value in the corresponding window in the image, the linear hidden crack feature in the image is enhanced.

[0045] In this implementation, the third gray value is obtained by processing the F gray responses of the third pixel point in the K directions; the linear hidden crack feature in the image can be enhanced.

[0046] In order to increase the contrast of the image, the gray value of each pixel point in the third image can also be normalized. In a possible implementation, after the second image is subjected to the hidden crack feature enhancement processing to obtain the third image, the method further includes: performing normalization processing on the gray value of each pixel point in the third image to obtain a fourth image; and the utilizing the trained defect recognition model to perform defect recognition on the third image and outputting the defect recognition result includes: utilizing the trained defect recognition model to perform defect recognition on the fourth image and outputting the defect recognition result.

[0047] The gray value of each pixel point in the third image after normalization processing and the gray value before normalization processing satisfy the following formula:

[0048]

[0049] wherein R' represents the gray value of each pixel point in the third image after normalization processing, R represents the gray value of the pixel point before normalization processing, R mean represents the average value of the gray value of each pixel point in the third image, R std represents the standard deviation of the gray value of each pixel point in the third image. It should be understood that the pixel points at the same position in the fourth image and the third image are one-to-one corresponding. R' can be the gray value of any pixel point in the fourth image, and R is the gray value of the pixel point in the third image corresponding to the any pixel point.

[0050] In this implementation, by normalizing the third image, the contrast of the gray value of each pixel point in the image can be enhanced, thereby enhancing the hidden crack feature.

[0051] In a possible implementation, after the second image is subjected to the hidden crack feature enhancement processing to obtain the third image, the method further includes: eliminating the grid lines in the third image and the boundary of the third image to obtain a fourth image; and the utilizing the trained defect recognition model to perform defect recognition on the third image and outputting the defect recognition result includes: utilizing the trained defect recognition model to perform defect recognition on the fourth image and outputting the defect recognition result.

[0052] In this implementation, by eliminating the grid lines in the third image and the boundary of the third image, the grid lines or the boundary can be avoided to be recognized as hidden cracks.

[0053] In a possible implementation, the defect recognition model is InceptionV4.

[0054] The InceptionV4 is a convolutional neural network. The defect recognition model can be obtained by training the InceptionV4 using positive samples and negative samples as training data.

[0055] In this implementation, the defect recognition model obtained by training the InceptionV4 can accurately identify the hidden crack defects.

[0056] In one possible implementation, the defect recognition model includes a Stem module, an Inception-A module, a Reduction-A module, and an Inception-C module. The second image (i.e., input) is first processed by the Stem module, and then sequentially processed by the Inception-A module, the Reduction-A module, and the Inception-C module, and finally processed by the Softmax classification.

[0057] The first image can be a 300x300 image. Considering the rotation invariance of the convolutional neural network, the model is transformed before the EL image is input into the model to increase the number of training samples. In this way, a defect recognition model with good effect can be trained using fewer training samples.

[0058] In a second aspect, the embodiments of the present application provide a method for detecting hidden cracks in a photovoltaic module. The method includes: performing hidden crack feature enhancement processing on a fifth image to obtain a sixth image; the fifth image is an electroluminescence (EL) image obtained by detecting the photovoltaic module; the hidden crack feature enhancement processing includes: obtaining F gray scale responses of a fourth pixel point in the fifth image in K directions, and adjusting the gray scale value of the fourth pixel point to a fourth gray scale value; the fourth gray scale value is obtained using the F gray scale responses; the gray scale response of the fourth pixel point in a second direction is used to determine whether there is a hidden crack in the second direction; the second direction is included in the K directions; K is an integer greater than 1; F is an integer equal to or greater than K; performing defect recognition on the sixth image using a trained defect recognition model, and outputting defect recognition information; the defect recognition information indicates whether there is a hidden crack in the photovoltaic module.

[0059] In the embodiments of the present application, the fifth image is first processed for hidden crack feature enhancement to obtain the sixth image, and then the trained defect recognition model is used to perform defect recognition on the sixth image, which can improve the accuracy of defect recognition.

[0060] In a possible implementation, the obtaining the F gray scale responses of the fourth pixel point in the fifth image in K directions includes: obtaining a second gray scale matrix; the second gray scale matrix is obtained by point multiplication of a sliding matrix and a second sub-matrix, a first element in the sliding matrix is not zero, a second element in the sliding matrix is zero, a direction of a first ray corresponding to the first element is included in the K directions, the first ray is a ray in the sliding matrix starting from a center element and passing through the first element, a direction of a second ray corresponding to the second element is not included in the K directions, the second ray is a ray in the sliding matrix starting from the center element and passing through the second element, a center element of the second sub-matrix is the fourth pixel point and the second sub-matrix is a sub-matrix of the fifth image; the F gray scale responses of the fourth pixel point in the K directions are obtained according to the second gray scale matrix; K rays starting from the center element in the second gray scale matrix correspond to the K directions one by one, the second direction corresponds to a fourth ray in the K rays, the fourth ray passes through two or more elements in the second gray scale matrix, and the gray scale response of the fourth pixel point in the second direction includes the fourth ray passing through the two or more elements in the second gray scale matrix.

[0061] In this implementation, the F gray scale responses of the fourth pixel point in the K directions can be obtained more accurately according to the second gray scale matrix.

[0062] In a possible implementation, the directions of the K rays starting from the center element in the sliding matrix correspond to the K directions one by one, and an included angle between any two adjacent rays in the K rays starting from the center element in the sliding matrix is a target angle.

[0063] In this implementation, the included angle between any two adjacent rays in the K rays is the target angle, so that the gray scale responses in each direction are detected.

[0064] In a possible implementation, the second gray scale matrix is an M-row and M-column matrix, the fourth ray passes through L elements in the second gray scale matrix, M is an odd number greater than 1, and L is an integer greater than 0 and not greater than M; after the F gray scale responses of the fourth pixel point in the K directions are obtained according to the second gray scale matrix, the method further includes: processing the F gray scale responses of the fourth pixel point in the K directions to obtain the fourth gray scale value. F=L*K. The fourth pixel point corresponds to L gray scale responses in each of the K directions. In some embodiments, the way in which the fourth gray scale value is processed by the crack detection device can be similar to the way in which the third gray scale value is processed.

[0065] In a possible implementation, after the fifth image is subjected to the hidden crack feature enhancement processing to obtain a sixth image, the method further includes: eliminating a grid line in the sixth image and a boundary of the sixth image to obtain a seventh image; and the defect recognition information output by the defect recognition model trained is obtained by performing defect recognition on the seventh image.

[0066] In this implementation, eliminating the grid line in the sixth image and the boundary of the sixth image can avoid identifying the grid line or the boundary as a hidden crack.

[0067] In a third aspect, an embodiment of the present application provides a hidden crack detection device, which includes: a feature extraction processing unit, configured to perform hidden crack feature extraction processing on a first image to obtain a second image; the first image is an electroluminescence (EL) image obtained by detecting a photovoltaic module, and the hidden crack feature extraction processing includes: adjusting a gray value of a first pixel point to a first gray value, and adjusting a gray value of a second pixel point to a second gray value; the first pixel point and the second pixel point are both included in the first image, a first eigenvalue of a first Hessian matrix corresponding to the first pixel point satisfies a target condition, a second eigenvalue of a second Hessian matrix corresponding to the second pixel point does not satisfy the target condition, the first gray value is obtained by using the first eigenvalue, and the second gray value is different from the first gray value; a defect recognition unit, configured to perform defect recognition on the second image by using a defect recognition model trained to obtain a defect recognition result; the defect recognition result indicates that the photovoltaic module has a hidden crack or does not have a hidden crack; and an output unit, configured to output the defect recognition result; the defect recognition result indicates that the photovoltaic module has a hidden crack or does not have a hidden crack.

[0068] In the embodiment of the present application, the gray value of the first pixel point is adjusted to the first gray value, and the gray value of the second pixel point is adjusted to the second gray value; the hidden crack feature in the first image can be accurately extracted, so that the accuracy of identifying the hidden crack defect in the photovoltaic module is improved.

[0069] In a possible implementation, the first eigenvalue satisfying the target condition includes that a third eigenvalue included in the first eigenvalue is less than or equal to 0, the third eigenvalue is an eigenvalue with a larger absolute value in two eigenvalues included in the first eigenvalue, and the second eigenvalue not satisfying the target condition includes that a fourth eigenvalue included in the second eigenvalue is greater than 0, the fourth eigenvalue is an eigenvalue with a larger absolute value in two eigenvalues included in the second eigenvalue.

[0070] In a possible implementation, the feature extraction processing unit is further configured to construct the first Hessian matrix corresponding to the first pixel point and construct the second Hessian matrix corresponding to the second pixel point; and process the first Hessian matrix to obtain the first eigenvalue and process the second Hessian matrix to obtain the second eigenvalue.

[0071] In a possible implementation, the feature extraction processing unit is specifically configured to perform hidden crack feature extraction processing on the first image by using the formula (5).

[0072] In a possible implementation, the hidden crack detection apparatus further includes a feature enhancement processing unit configured to perform hidden crack feature enhancement processing on the second image to obtain a third image; the hidden crack feature enhancement processing includes obtaining F gray scale responses of a third pixel point in the second image in K directions, and adjusting a gray scale value of the third pixel point to a third gray scale value; the third gray scale value is obtained by using the F gray scale responses, and a gray scale response of the third pixel point in a first direction is used to determine whether there is a hidden crack in the first direction of the third pixel point, the first direction is included in the K directions, K is an integer greater than 1, and F is an integer equal to or greater than K; and the defect recognition unit is specifically configured to perform defect recognition on the third image by using the trained defect recognition model to obtain the defect recognition result.

[0073] In a possible implementation, the feature enhancement processing unit is specifically configured to obtain a first gray scale matrix; the first gray scale matrix is obtained by point multiplication of a sliding matrix and a first sub-matrix, a first element in the sliding matrix is not zero, a second element in the sliding matrix is zero, a direction of a first ray corresponding to the first element is included in the K directions, the first ray is a ray in the sliding matrix that starts from a center element and passes through the first element, a direction of a second ray corresponding to the second element is not included in the K directions, the second ray is a ray in the sliding matrix that starts from the center element and passes through the second element, a center element of the first sub-matrix is the third pixel point, and the first sub-matrix is a sub-matrix of the second image; the F gray scale responses of the third pixel point in the K directions are obtained according to the first gray scale matrix; directions of K rays in the first gray scale matrix that start from the center element correspond to the K directions one by one, the first direction corresponds to a third ray in the K rays, the third ray passes through two or more elements in the first gray scale matrix, and the gray scale response of the third pixel point in the first direction includes the third ray passing through the two or more elements in the first gray scale matrix.

[0074] In a possible implementation, directions of K rays starting from the center element in the sliding matrix correspond to the K directions one by one, and an included angle between any two adjacent rays in the K rays starting from the center element in the sliding matrix is the target angle.

[0075] In a possible implementation, the first gray scale matrix is an M-row and M-column matrix, the third ray passes through L elements in the first gray scale matrix, the M is an odd number greater than 1, and the L is an integer greater than 0 and not greater than the M; and the feature enhancement processing unit is further configured to obtain the third gray scale value by processing the F gray scale responses of the third pixel point in the K directions.

[0076] In a possible implementation, the feature enhancement processing unit is further configured to perform normalization processing on gray scale values of all pixel points in the third image to obtain a fourth image; and the defect identification unit is specifically configured to perform defect identification on the fourth image by using the trained defect identification model, to obtain the defect identification result.

[0077] In a possible implementation, the hidden crack detection apparatus further includes an interference elimination unit configured to eliminate a grid line in the third image and a boundary of the third image to obtain a fourth image; and the defect identification unit is specifically configured to perform defect identification on the fourth image by using the trained defect identification model, and output the defect identification result.

[0078] In a possible implementation, the defect identification model is InceptionV4.

[0079] In a possible implementation, the defect identification model includes a Stem module, an Inception-A module, a Reduction-A module, and an Inception-C module; wherein the second image (input) first passes through the Stem module, and then sequentially passes through the Inception-A module, the Reduction-A module, and the Inception-C module, and finally passes through Softmax classification.

[0080] As to the technical effects brought by the third aspect or various possible implementations, reference can be made to the introduction of the technical effects of the first aspect or corresponding implementations.

[0081] In a fourth aspect, an example of the present application provides a hidden crack detection device, comprising: a hidden crack enhancement unit, configured to perform hidden crack feature enhancement processing on a fifth image to obtain a sixth image; the fifth image is an electroluminescence (EL) image obtained by detecting a photovoltaic module; the hidden crack feature enhancement processing comprises: obtaining F gray scale responses of a fourth pixel point in the fifth image in K directions, and adjusting a gray scale value of the fourth pixel point to a fourth gray scale value; the fourth gray scale value is obtained by using the F gray scale responses; a gray scale response of the fourth pixel point in a second direction is used to determine whether there is a hidden crack in the second direction; the second direction is included in the K directions; K is an integer greater than 1; F is an integer equal to or greater than K; a defect processing unit, configured to perform defect recognition on the sixth image by using a trained defect recognition model to obtain defect recognition information, the defect recognition information indicating whether there is a hidden crack in the photovoltaic module; and an output unit, configured to output the defect recognition information.

[0082] In a possible implementation, the hidden crack enhancement unit is specifically configured to obtain a second gray scale matrix; the second gray scale matrix is obtained by point multiplication of a sliding matrix and a second sub-matrix; a first element in the sliding matrix is not zero, a second element in the sliding matrix is zero, a direction of a first ray corresponding to the first element is included in the K directions, the first ray is a ray in the sliding matrix starting from a center element and passing through the first element, a direction of a second ray corresponding to the second element is not included in the K directions, the second ray is a ray in the sliding matrix starting from the center element and passing through the second element, a center element of the second sub-matrix is the fourth pixel point, and the second sub-matrix is a sub-matrix of the fifth image; the F gray scale responses of the fourth pixel point in the K directions are obtained according to the second gray scale matrix; K rays starting from a center element in the second gray scale matrix correspond to the K directions in a one-to-one manner, the second direction corresponds to a fourth ray in the K rays, the fourth ray passes through two or more elements in the second gray scale matrix, and the gray scale response of the fourth pixel point in the second direction includes the fourth ray passing through the two or more elements in the second gray scale matrix.

[0083] In a possible implementation, directions of K rays starting from a center element in the sliding matrix correspond to the K directions in a one-to-one manner, and an included angle between any two adjacent rays in the K rays starting from the center element in the sliding matrix is a target angle.

[0084] In a possible implementation, the second gray scale matrix is an M-row and M-column matrix, the fourth ray passes through L elements in the second gray scale matrix, M is an odd number greater than 1, and L is an integer greater than 0 and not greater than M; and the hidden crack enhancement unit is further configured to obtain the fourth gray scale value by processing the F gray scale responses of the fourth pixel point in the K directions.

[0085] In a possible implementation, the hidden crack detection apparatus further includes an interference processing unit, configured to eliminate the grid lines in the sixth image and the boundary of the sixth image to obtain a seventh image; and the defect processing unit is specifically configured to perform defect recognition on the seventh image by using the trained defect recognition model to obtain the defect recognition information.

[0086] As to the technical effects brought by the fourth aspect or various possible implementations, reference can be made to the introduction of the technical effects of the second aspect or corresponding implementations.

[0087] In a fifth aspect, the present application provides a chip, which includes a processing circuit and an interface circuit, the interface circuit is configured to acquire or output data; and the processing circuit is configured to execute the method as shown in the first aspect or any possible implementation of the first aspect.

[0088] In a sixth aspect, the present application provides a chip, which includes a processing circuit and an interface circuit, the interface circuit is configured to acquire or output data; and the processing circuit is configured to execute the method as shown in the second aspect or any possible implementation of the second aspect.

[0089] In a seventh aspect, the embodiments of the present application provide an electronic device, which includes a processor, a memory, an input device and an output device, the memory is configured to store code; the processor is configured to read the code stored in the memory to execute the method as shown in the first aspect or any possible implementation of the first aspect, the input device is configured to obtain an image to be subjected to defect recognition, and the output device is configured to output the defect recognition result obtained by the processor processing the image.

[0090] In an eighth aspect, the embodiments of the present application provide an electronic device, which includes a processor, a memory, an input device and an output device, the memory is configured to store code; the processor is configured to read the code stored in the memory to execute the method as shown in the second aspect or any possible implementation of the second aspect, the input device is configured to obtain an image to be subjected to defect recognition, and the output device is configured to output the defect recognition information obtained by the processor processing the image.

[0091] In a ninth aspect, an embodiment of the present application provides a computer program product, which comprises program instructions, and the program instructions, when executed by a processor, cause the processor to perform the method in the first aspect or any possible implementation manner of the first aspect.

[0092] In a tenth aspect, an embodiment of the present application provides a computer program product, which comprises program instructions, and the program instructions, when executed by a processor, cause the processor to perform the method in the first aspect or any possible implementation manner of the first aspect.

[0093] In an eleventh aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program comprises program instructions, and the program instructions, when executed by a processor, cause the processor to perform the method in the first aspect or any possible implementation manner of the first aspect.

[0094] In a twelfth aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program comprises program instructions, and the program instructions, when executed by a processor, cause the processor to perform the method in the second aspect or any possible implementation manner of the second aspect.

[0095] In a thirteenth aspect, an embodiment of the present application provides a hidden crack detection system for a photovoltaic module, comprising: a photovoltaic power station management device, a data collector, a control device, a plurality of inverters, a plurality of photovoltaic module strings, a hidden crack detection device, and an image acquisition device; the image acquisition device is configured to move according to a control instruction from the control device; the image acquisition device is configured to send pose information to the control device; the pose information is used to indicate the pose of the image acquisition device; the control device is configured to send a back-feeding current request to the data collector; the back-feeding current request comprises an identifier of a target inverter; the target inverter is an inverter coupled to a target photovoltaic module string in the plurality of inverters, and the target photovoltaic module string belongs to a photovoltaic module string photographed by the image acquisition device at the pose; the data collector is configured to send the back-feeding current request to the photovoltaic power station management device; the photovoltaic power station management device is configured to send a back-feeding current instruction to the target inverter; the target inverter is configured to apply a back-feeding current to the target photovoltaic module string after receiving the back-feeding current instruction; the image acquisition device is configured to acquire an electroluminescence (EL) image of the target photovoltaic module string to obtain a first image; the hidden crack detection device is configured to obtain the first image and perform hidden crack detection using the first image to obtain a defect identification result; the defect identification result indicates whether the target photovoltaic module has a hidden crack or not, and the target photovoltaic module is included in the target photovoltaic module string.

[0096] In a possible implementation, the hidden crack detection device is configured to perform a hidden crack feature extraction process on the first image to obtain a second image; the hidden crack feature extraction process includes adjusting a gray value of a first pixel point to a first gray value and adjusting a gray value of a second pixel point to a second gray value; the first pixel point and the second pixel point are both included in the first image, a first eigenvalue of a first Hessian matrix corresponding to the first pixel point satisfies a target condition, a second eigenvalue of a second Hessian matrix corresponding to the second pixel point does not satisfy the target condition, the first gray value is obtained by using the first eigenvalue, and the second gray value is different from the first gray value; and the trained defect recognition model is used to perform defect recognition on the second image to output the defect recognition result.

[0097] In a possible implementation, the image acquisition device is a drone carrying an infrared camera, and the control device is a remote controller used to control the drone, and the control device is further configured to determine a photovoltaic string photographed by the image acquisition device according to the pose information to obtain the target photovoltaic string. BRIEF DESCRIPTION OF DRAWINGS

[0098] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the background art, the drawings needed to be used in the embodiments of the present application or the background art will be described below.

[0099] Figure 1 A schematic diagram of a photovoltaic module hidden crack detection system provided by an embodiment of the present application;

[0100] Figure 2 An interactive flowchart of a photovoltaic module hidden crack detection method provided by an embodiment of the present application;

[0101] Figure 3 A flowchart of a photovoltaic module hidden crack detection method provided by an embodiment of the present application;

[0102] Figure 4 A comparison schematic diagram of a first image and a second image provided by an embodiment of the present application;

[0103] Figure 5 A normal image of a single crystal cell, a hidden crack defect image of a single crystal cell, a normal image of a photovoltaic module, a hidden crack defect image of a photovoltaic module, and a gray value distribution diagram corresponding to each image are shown;

[0104] Figure 6 Another flowchart of a photovoltaic module hidden crack detection method provided by an embodiment of the present application;

[0105] Figure 7 A comparison schematic diagram of a second image and a third image provided by an embodiment of the present application;

[0106] Figure 8 Another flow chart of a method for detecting hidden cracks in a photovoltaic module is provided in embodiments of the present application;

[0107] Figure 9 Another flow chart of a method for detecting hidden cracks in a photovoltaic module is provided in embodiments of the present application;

[0108] Figure 10 Another flow chart of a method for detecting hidden cracks in a photovoltaic module is provided in embodiments of the present application;

[0109] Figure 11 A method for enhancing hidden crack features of an image is provided in embodiments of the present application;

[0110] Figure 12 A schematic diagram of a first gray matrix and K rays in the first gray matrix is provided in embodiments of the present application;

[0111] Figure 13 A schematic diagram of a sliding matrix and K rays in the sliding matrix is provided in embodiments of the present application;

[0112] Figure 14 A schematic diagram of a second image and a first sub-matrix is provided in embodiments of the present application;

[0113] Figure 15 A flow chart of a training method is provided in embodiments of the present application;

[0114] Figure 16 A structural schematic diagram of a defect recognition model is provided in embodiments of the present application;

[0115] Figure 17 A structural schematic diagram of a hidden crack detection device is provided in embodiments of the present application;

[0116] Figure 18 A structural schematic diagram of another hidden crack detection device is provided in embodiments of the present application;

[0117] Figure 19 A block diagram of a partial structure of an electronic device is provided in embodiments of the present application;

[0118] Figure 20 A structural schematic diagram of another electronic device 200 is provided in embodiments of the present application. DETAILED DESCRIPTION

[0119] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described below with reference to the accompanying drawings.

[0120] The terms "first" and "second" and the like in the description and in the claims of this application are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. The terms "comprises", "comprising", "includes", "including" and the like are used for indicating the inclusion of elements, steps, processes, etc. without precluding the presence or addition of one or more other elements, steps, processes, etc. The term "and / or", within the meaning of this document, refers to and encompasses any and all possible combinations of one or more of the associated items, for example, the case of "A and / or B" is a case where A is present alone, B is present alone, or both A and B are present. The term "a plurality of" refers to two or more.

[0121] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase "in an embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily all directed to the same embodiment, or to a single alternative embodiment. It is appreciated that one or more of the features, structures, or characteristics described in connection with an embodiment can be included in at least one implementation of the application.

[0122] As mentioned in the background, there is a current need to research a scheme that can accurately identify hidden crack defects in photovoltaic modules. With the rapid development of computer technology, deep learning has achieved large-scale applications in computer vision, natural language processing, and other fields due to its unique advantages. A theoretically feasible scheme for identifying hidden crack defects in photovoltaic modules is to train a defect recognition model capable of identifying hidden crack defects in EL images using a large number of EL image defect samples (i.e., training samples). The complexity of EL images necessitates training the defect recognition model with even more EL image defect samples to ensure the accuracy of the model in identifying hidden crack defects. Since EL image defect samples are often difficult to collect, this scheme is not suitable for training a defect recognition model capable of identifying hidden crack defects in EL images in practical applications. This application provides a scheme that first performs hidden crack feature extraction and / or hidden crack feature enhancement processing on EL images, and then uses the processed EL images to train a defect recognition model. By performing hidden crack feature extraction and / or hidden crack feature enhancement processing on EL images, the hidden crack features in EL images can be effectively extracted or enhanced, thereby reducing the complexity of EL images. Therefore, the solution provided in this application can train a defect recognition model that can effectively identify hidden cracks using only a small number of EL image defect samples. The principle of the solution for identifying hidden cracks in photovoltaic modules provided in this application is mainly as follows: first, the EL image undergoes hidden crack feature extraction and / or hidden crack feature enhancement processing; then, the trained defect recognition model is used to identify defects in the processed EL image, and the defect recognition result is output. The following section introduces a photovoltaic module hidden crack detection system to which the solution for identifying hidden cracks in photovoltaic modules provided in this application can be applied.

[0123] Figure 1 This is a schematic diagram of a photovoltaic module microcrack detection system provided in an embodiment of this application. Figure 1 As shown, the photovoltaic module microcrack detection system includes: photovoltaic power plant management equipment, a data acquisition unit, a control device, multiple inverters, multiple photovoltaic strings, a microcrack detection device, and an image acquisition device. The photovoltaic power plant management equipment can be coupled to the data acquisition unit. The photovoltaic power plant management equipment is coupled to multiple inverters. The multiple inverters are coupled to multiple photovoltaic strings. The control device is coupled to both the image acquisition device and the data acquisition unit. In this application, coupling can refer to wired or wireless connections.

[0124] The photovoltaic power station management device can be a computer device (e.g., a computer) running a power station management system, which can control the inverters and receive information from the data collectors. The power station management system can remotely monitor the operating status of the inverters, aggregate the direct current of the inverters and combiner boxes, analyze the data, and timely discover faults of the inverters, combiner boxes, and other components. The data collectors can be hardware with communication functions with the control devices and the photovoltaic power station management device. For example, the data collectors communicate with the photovoltaic power station management device and the control devices through WiFi, respectively. The photovoltaic module hidden crack detection system can include a plurality of data collectors, Figure 1 Only one is shown. The control device can be a remote controller that controls the movement of the image acquisition device. Each photovoltaic module string can include a plurality of photovoltaic modules. Each inverter is connected to one or more photovoltaic module strings. The inverter is a converter that converts direct current into alternating current. Each inverter converts the direct current provided by the photovoltaic module strings connected thereto into alternating current and can also apply a reverse current to the photovoltaic module strings connected thereto to enable electroluminescence of the photovoltaic module strings. The hidden crack detection device can be a notebook computer, a desktop computer, a server, or other device with data processing capability. The image acquisition device can be a drone carrying an infrared camera, a vehicle carrying an infrared camera, or other device carrying an infrared camera. The image acquisition device can move under the control of the control device and acquire images using the infrared camera. In one possible implementation, the photovoltaic power station management device is connected to the data collectors through a wired manner (e.g., optical fiber). The photovoltaic power station management device is connected to each inverter through a wired manner, respectively. The control device can be connected to the data collectors through a wireless manner (e.g., WiFi). The control device can be connected to the image acquisition device through a wireless manner (e.g., WiFi). The functions of the devices in the photovoltaic module hidden crack detection system are described below.

[0125] The image acquisition device is configured to move according to the control instruction from the control device and send pose information to the control device.

[0126] The control device is configured to send a reverse current request to the data collector.

[0127] The data collector is configured to send a reverse current request to the photovoltaic power station management device.

[0128] The photovoltaic power station management device is configured to send a reverse current instruction to the target inverter.

[0129] The target inverter is configured to apply a reverse current to the target photovoltaic module string after receiving the reverse current instruction.

[0130] The image acquisition device is configured to acquire an electroluminescence (EL) image of the target photovoltaic module string to obtain a first image.

[0131] The hidden crack detection device is configured to obtain a first image and perform hidden crack detection on the first image to obtain a defect identification result. The defect identification result indicates whether the target photovoltaic module has a hidden crack or not. The target photovoltaic module is included in a target photovoltaic module string.

[0132] The following describes operations of each device in the photovoltaic module hidden crack detection system in the process of identifying hidden crack defects in the photovoltaic module. Figure 2 An interactive flowchart of a photovoltaic module hidden crack detection system is provided in an embodiment of the present application. As shown in the figure, the interactive flowchart includes the following steps. Figure 2

[0133] 201. The control device sends a control instruction to the image acquisition device.

[0134] The control instruction is used to control the image acquisition device to move. For example, a user can control the image acquisition device to move to a target position by using a control device (e.g., a remote controller) so that the image acquisition device can acquire an EL image at the target position by using an infrared camera.

[0135] 202. The image acquisition device moves according to the control instruction from the control device and sends pose information to the control device.

[0136] The pose information is used to indicate a target pose of the image acquisition device. The target pose can represent a current position and attitude of the image acquisition device. The pose information can include the current position and attitude of the image acquisition device.

[0137] 203. The control device determines a target inverter according to the pose information and sends a reverse current request to the data collector.

[0138] The reverse current request includes an identifier of the target inverter. The target inverter is an inverter that is coupled to a target photovoltaic module string among a plurality of inverters, and the target photovoltaic module string belongs to a photovoltaic module string that can be photographed by the image acquisition device at the target pose.

[0139] The control device can determine the target inverter according to the pose information in the following manner: the control device determines a photovoltaic module string that can be photographed by the image acquisition device at the target pose according to the pose information, to obtain a target photovoltaic module string; and the control device obtains a target inverter that is connected to the target photovoltaic module string. For example, the control device stores a pose-photovoltaic module string correspondence table, which includes a correspondence between a target pose and a photovoltaic module string that can be photographed by the image acquisition device at the target pose. In this example, the control device can determine the photovoltaic module string corresponding to the target pose by using the correspondence table.

[0140] 204. The data collector sends the reverse current request to a photovoltaic power plant management device.

[0141] ​205. The photovoltaic power station management device sends a back-feeding current instruction to the target inverter.

[0142] 206. The target inverter applies a back-feeding current to the target photovoltaic string after receiving the back-feeding current instruction.

[0143] 207. The image acquisition device acquires an electroluminescence (EL) image of the target photovoltaic string to obtain a first image.

[0144] 208. The hidden crack detection device obtains the first image and performs hidden crack detection using the first image to obtain a defect identification result.

[0145] The defect identification result indicates whether the target photovoltaic module has a hidden crack or not, and the target photovoltaic module is included in the target photovoltaic string. In one possible implementation, the image acquisition device stores the acquired EL image in a storage card, such as a secure digital memory card (SD), and copies the EL image in the storage card to the hidden crack detection device. The defect identification result can be regarded as a fault diagnosis report. In some embodiments, the fault diagnosis software running on the hidden crack detection device performs hidden crack detection using the first image to obtain the defect identification result, i.e., the fault diagnosis report. The hidden crack detection device can output the defect identification result.

[0146] In one possible implementation, the hidden crack detection device is specifically configured to perform hidden crack feature extraction processing on the first image to obtain a second image, and the hidden crack feature extraction processing includes: adjusting a gray value of a first pixel point to a first gray value, and adjusting a gray value of a second pixel point to a second gray value; the first pixel point and the second pixel point are both included in the first image, a first eigenvalue of a first Hessian matrix corresponding to the first pixel point satisfies a target condition, a second eigenvalue of a second Hessian matrix corresponding to the second pixel point does not satisfy the target condition, the first gray value is obtained by using the first eigenvalue, and the second gray value is different from the first gray value; and a defect identification model trained is used to perform defect identification on the second image, and the defect identification result is output.

[0147] In the embodiments of the present application, the control device determines the photovoltaic string that can be currently photographed by the image acquisition device according to the pose information fed back by the image acquisition device, and sends a back-feeding current request to the photovoltaic power station management device, so that the photovoltaic power station management device controls the corresponding photovoltaic string to realize electroluminescence through the inverter; and the EL image of the photovoltaic string at any position can be quickly acquired.

[0148] Figure 2Steps 201 to 207 in FIG. 2 describe an example of acquiring an EL image. It should be understood that the devices in the photovoltaic module hidden crack detection system can also acquire an EL image in other ways, which are not limited in the present application. The focus of the present application is how to identify the hidden crack defects in the photovoltaic module. The scheme provided by the embodiments of the present application for identifying the hidden cracks in the photovoltaic module is described below with reference to the accompanying drawings.

[0149] Figure 3 A flow chart of a photovoltaic module hidden crack detection method provided by the embodiments of the present application is shown in FIG. 3. As shown in FIG. 3, the method comprises the following steps. Figure 3

[0150] 301. The hidden crack detection device performs hidden crack feature extraction processing on the first image to obtain a second image.

[0151] The hidden crack detection device can be a notebook computer, a desktop computer, a server, or other devices with data processing capability. The first image is an electroluminescence (EL) image obtained by detecting the photovoltaic module. The hidden crack feature extraction processing comprises adjusting the gray value of a first pixel point to a first gray value and adjusting the gray value of a second pixel point to a second gray value. The first pixel point and the second pixel point are both included in the first image. A first eigenvalue of a first Hessian matrix corresponding to the first pixel point satisfies a target condition. The first eigenvalue satisfying the target condition can be that a third eigenvalue included in the first eigenvalue is less than or equal to 0. The third eigenvalue is the eigenvalue with a larger absolute value among two eigenvalues included in the first eigenvalue. A second eigenvalue of a second Hessian matrix corresponding to the second pixel point does not satisfy the target condition. The second eigenvalue not satisfying the target condition can be that a fourth eigenvalue included in the second eigenvalue is greater than 0. The fourth eigenvalue is the eigenvalue with a larger absolute value among two eigenvalues included in the second eigenvalue. The first gray value is obtained by using the first eigenvalue. The second gray value is different from the first gray value. The first pixel point can be understood as any pixel point in the first image corresponding to a hidden crack defect, and the second pixel point can be understood as any pixel point in the first image not corresponding to a hidden crack defect. The hidden crack detection device can adjust the gray value of each pixel point in the first image according to whether the eigenvalue of the Hessian matrix corresponding to each pixel point satisfies the target condition.

[0152] In one possible implementation, the hidden crack feature extraction processing on the first image is performed by using the following formula:

[0153]

[0154] wherein R b = λ1 / λ2, S = λ1 2 + λ2 2 ​, λ1, λ2 are eigenvalues of the Hessian matrix corresponding to any pixel point u in the first image (for example, the first eigenvalue or the second eigenvalue), and β and c are fixed parameters for controlling sensitivity, is the gray value of any pixel point u in the first image after the hidden crack feature extraction. β and c can be configured according to actual needs. It should be understood that the hidden crack detection device can also use a formula similar to formula (10) to perform hidden crack feature extraction processing on the first image. For example, when λ2 is less than or equal to 0, g is a fixed parameter; when λ2 is greater than 0, is equal to 0. In the above formula (10), λ2 can be an eigenvalue with a larger absolute value among the eigenvalues of the Hessian matrix corresponding to any pixel point in the first image. As can be seen from formula (10), if the eigenvalue with a larger absolute value among the two eigenvalues of the Hessian matrix corresponding to any pixel point u (for example, the first pixel point) in the first image is less than or equal to 0, the gray value of the any pixel point u is adjusted to (corresponding to the first gray value). As can be seen from formula (10), if the eigenvalue with a larger absolute value among the two eigenvalues of the Hessian matrix corresponding to any pixel point u (for example, the second pixel point) in the first image is greater than 0, the gray value of the any pixel point u is adjusted to 0 (corresponding to the second gray value). The pixel point whose gray value is adjusted to can be regarded as a pixel point corresponding to a hidden crack defect that is preliminarily screened out, and the pixel point whose gray value is adjusted to 0 can be regarded as a pixel point not corresponding to a hidden crack defect that is preliminarily screened out. As can be seen, formula (10) can accurately distinguish the pixel points corresponding to the hidden crack defects and the pixel points not corresponding to the hidden crack defects. In actual application, after the eigenvalues of the Hessian matrix corresponding to any pixel point in the first image are obtained, the hidden crack detection device can adjust the gray value of the any pixel point by using the above formula (10). It can be understood that after the gray values of each pixel point in the first image are processed according to the above formula (10), a second image can be obtained.

[0155] Figure 4 is a comparison schematic diagram of a first image and a second image provided by an embodiment of the present application. Figure 4 In the figure, 401 represents the first image, and 402 represents the second image. By comparing the first image and the second image, it can be seen that after the first image passes through the feature extraction link (i.e., the hidden crack feature extraction processing), the hidden crack feature becomes more obvious. Figure 5The normal image of the single crystal cell piece, the hidden crack defect image of the single crystal cell piece, the normal image of the photovoltaic module (for example, a polycrystalline silicon cell piece), the hidden crack defect image of the photovoltaic module, and the corresponding gray scale distribution diagram of each image are shown. The normal image of the single crystal cell piece refers to the EL image of the single crystal cell piece without hidden crack defects. The hidden crack defect image of the single crystal cell piece refers to the EL image of the single crystal cell piece with hidden crack defects. The normal image of the photovoltaic module refers to the EL image of the photovoltaic module without hidden crack defects. The hidden crack defect image of the photovoltaic module refers to the EL image of the photovoltaic module with hidden crack defects. Figure 5 In the figure, (a1) represents the normal image of the single crystal cell piece, (b1) represents the hidden crack defect image of the single crystal cell piece, (c1) represents the normal image of the photovoltaic module, (d1) represents the hidden crack defect image of the photovoltaic module, (a2) represents the gray scale distribution diagram corresponding to the normal image of the single crystal cell piece, (b2) represents the gray scale distribution diagram corresponding to the hidden crack defect image of the single crystal cell piece, (c2) represents the gray scale distribution diagram corresponding to the normal image of the photovoltaic module, and (d2) represents the gray scale distribution diagram corresponding to the hidden crack defect image of the photovoltaic module. It can be seen from the figure that the numerical distribution in the hidden crack defect image of the single crystal cell piece and the hidden crack defect image of the photovoltaic module has obvious difference, and the hidden crack characteristics in the hidden crack defect image of the photovoltaic module become very inconspicuous due to the influence of the crystal node. Therefore, it is necessary to perform hidden crack feature extraction processing and / or hidden crack feature enhancement processing on the EL of the photovoltaic module, and then accurately identify the hidden crack defects in the photovoltaic module. Figure 5 In the figure, (a1) represents the normal image of the single crystal cell piece, (b1) represents the hidden crack defect image of the single crystal cell piece, (c1) represents the normal image of the photovoltaic module, (d1) represents the hidden crack defect image of the photovoltaic module, (a2) represents the gray scale distribution diagram corresponding to the normal image of the single crystal cell piece, (b2) represents the gray scale distribution diagram corresponding to the hidden crack defect image of the single crystal cell piece, (c2) represents the gray scale distribution diagram corresponding to the normal image of the photovoltaic module, and (d2) represents the gray scale distribution diagram corresponding to the hidden crack defect image of the photovoltaic module. It can be seen from the figure that the numerical distribution in the hidden crack defect image of the single crystal cell piece and the hidden crack defect image of the photovoltaic module has obvious difference, and the hidden crack characteristics in the hidden crack defect image of the photovoltaic module become very inconspicuous due to the influence of the crystal node. Therefore, it is necessary to perform hidden crack feature extraction processing and / or hidden crack feature enhancement processing on the EL of the photovoltaic module, and then accurately identify the hidden crack defects in the photovoltaic module.

[0156] 302, The hidden crack detection device performs defect recognition on the second image by using the trained defect recognition model, and outputs a defect recognition result.

[0157] The defect recognition result indicates whether the photovoltaic module has hidden cracks or not. For example, the defect recognition result is 1, which indicates that the photovoltaic module has hidden cracks. For another example, the defect recognition result is 0, which indicates that the photovoltaic module does not have hidden cracks.

[0158] The defect recognition model can be InceptionV4, and can also be InceptionV3, InceptionV2, InceptionV1, etc., which are not limited in the application. The defect recognition model can be obtained by training InceptionV4 by using positive samples (i.e., EL image defect samples) and negative samples as training samples. The method process of training the defect recognition model will be described later. The positive sample refers to a sample obtained by performing hidden crack feature extraction processing and / or hidden crack feature enhancement processing on an EL image containing hidden crack defects. The negative sample refers to a sample obtained by performing hidden crack feature extraction processing and / or hidden crack feature enhancement processing on an EL image not containing hidden crack defects.

[0159] In the embodiment of the present application, the gray value of the first pixel point is adjusted to the first gray value, and the gray value of the second pixel point is adjusted to the second gray value; the hidden crack feature in the first image can be accurately extracted, thereby improving the accuracy of identifying the hidden crack defect in the photovoltaic module.

[0160] Figure 6 Another flow chart of a method for detecting hidden cracks in a photovoltaic module is provided in the embodiment of the present application. Figure 6 The method flow in the embodiment of the present application is an improvement on the method flow in the embodiment of the present application. Figure 3 As shown in the embodiment of the present application, the method comprises the following steps. Figure 6

[0161] 601. The hidden crack detection device performs hidden crack feature extraction processing on the first image to obtain a second image.

[0162] The implementation manner of step 601 can be the same as that of step 301.

[0163] 602. The hidden crack detection device performs hidden crack feature enhancement processing on the second image to obtain a third image.

[0164] The hidden crack feature enhancement processing comprises obtaining F gray responses of a third pixel point in the second image in K directions, and adjusting the gray value of the third pixel point to a third gray value. The third gray value is obtained by using the F gray responses. The gray response of the third pixel point in the first direction is used to determine whether there is a hidden crack in the third pixel point in the first direction. The first direction is included in the K directions, and K is an integer greater than 1. F is an integer equal to or greater than K. Any direction in the K directions can be the direction of the ray passing through two or more pixel points in the second image with the third pixel point in the second image as the starting point (i.e. the end point). The third pixel point can be any pixel point in the second image. F can be equal to K. Alternatively, F can be an integer multiple of K. For example, F = L*K, and L is an integer greater than 1. It should be understood that the hidden crack feature enhancement processing can be: obtaining F gray responses of each pixel point in the second image in K directions; and then adjusting the gray value of each pixel point according to the F gray responses of each pixel point in K directions. The F gray responses of the third pixel point in the second image in each direction can be used to determine whether there is a hidden crack in the third pixel point in the direction. Since the hidden crack in the image usually has a linear feature, the F gray responses of the third pixel point in the K directions can further enhance the hidden crack feature. The implementation manner of the hidden crack feature enhancement processing on the second image will be described in detail later. The purpose of the hidden crack feature enhancement processing on the second image is to further highlight the hidden crack feature. Figure 7 A comparison diagram of the second image and the third image provided in the embodiment of the present application is shown.​Figure 7 In the figure, 701 represents the second image, and 702 represents the third image. By comparing the second image and the third image, it can be seen that the third image is obviously weaker than the second image in noise, and the most obvious part is the grid line and the place where the hidden crack defect exists.

[0165] 603. The defect recognition model trained is used to perform defect recognition on the third image, and a defect recognition result is output.

[0166] In one possible implementation, before performing step 603, the hidden crack detection apparatus can perform binarization processing on the third image; then, the defect recognition model trained is used to perform defect recognition on the third image after binarization processing, and a defect recognition result is output. Using the defect recognition model trained to perform defect recognition on the third image after binarization processing can reduce the complexity of the image processed by the defect recognition model. The binarization processing can be adjusting each gray value in the image that is less than a certain threshold (for example, 100, 200, etc.) to 0, and adjusting each gray value that is greater than or equal to the threshold to 255.

[0167] In the embodiments of the present application, the hidden crack feature extraction is first performed on the first image to obtain the second image, and then the hidden crack feature enhancement processing is performed on the second image to obtain the third image, which can enhance the hidden crack features in the effective enhanced image, thereby improving the accuracy of recognizing the hidden crack defects in the photovoltaic module.

[0168] Figure 8 Another photovoltaic module hidden crack detection method flowchart provided in the embodiments of the present application. Figure 8 The method flowchart in the present application is an improvement on the method flowchart in the prior art. Figure 6 As shown in FIG. 8, the method includes the following steps. Figure 8

[0169] 801. The hidden crack detection apparatus performs hidden crack feature extraction processing on the first image to obtain a second image.

[0170] The implementation of step 801 can be the same as that of step 601.

[0171] 802. The hidden crack detection apparatus performs hidden crack feature enhancement processing on the second image to obtain a third image.

[0172] The implementation of step 802 can be the same as that of step 602.

[0173] 803. The hidden crack detection apparatus eliminates the grid line in the third image and the boundary of the third image to obtain a fourth image.

[0174] ​In a possible implementation, the hidden crack detection apparatus can perform binarization processing on the fourth image before performing step 804; then, the trained defect recognition model is used to perform defect recognition on the binarized fourth image, and a defect recognition result is output.

[0175] 804. Perform defect recognition on the fourth image by using the trained defect recognition model, and output a defect recognition result.

[0176] Figure 8 In the method flow, 8001 represents the first image, 8002 represents the second image, 8003 represents the third image, and 8004 represents the fourth image. The hidden crack feature of the second image 8002 is more obvious than that of the first image 8001. The hidden crack feature of the third image 8003 is further enhanced compared with the second image 8002. The three longest white lines in the third image represent the grid lines. The fourth image 8004 is compared with the third image 8003, and the grid lines are eliminated.

[0177] In a possible implementation, the hidden crack detection apparatus can perform binarization processing on the fourth image before performing step 804; then, the trained defect recognition model is used to perform defect recognition on the binarized fourth image, and a defect recognition result is output.

[0178] Figure 8 Compared with the method flow in Figure 6 , the method flow in the present embodiment adds the operation of eliminating the grid lines and the boundaries in the image, so that the grid lines and the boundaries can be avoided from being recognized as hidden crack defects, thereby improving the accuracy of recognizing the hidden crack defects in the photovoltaic module.

[0179] Figure 3 、 Figure 6 and Figure 8 The method flow in each of the above embodiments includes the operation of performing hidden crack feature extraction processing on the first image. The present embodiment further provides another method flow of detecting hidden cracks in a photovoltaic module, which does not include the operation of performing hidden crack feature extraction processing on the image. The method flow is described below with reference to the accompanying drawings.

[0180] Figure 9 Another method flow chart of detecting hidden cracks in a photovoltaic module is provided in the present embodiment. Figure 9 The method flow in the present embodiment does not include the operation of performing hidden crack feature extraction processing on the image. As shown in Figure 9 , the method includes the following steps.

[0181] 901. The microcrack detection device performs microcrack feature enhancement processing on the fifth image to obtain the sixth image.

[0182] The fifth image mentioned above is an electroluminescent (EL) image obtained by detecting a photovoltaic module. The microcrack feature enhancement processing includes: obtaining F grayscale responses of the fourth pixel in the fifth image in K directions, and adjusting the grayscale value of the fourth pixel to a fourth grayscale value. The fourth grayscale value is obtained using the F grayscale responses. The grayscale response of the fourth pixel in the second direction is used to determine whether a microcrack exists in the fourth pixel in the second direction. The second direction is included in the K directions. K is an integer greater than 1, and F is an integer equal to or greater than K. The implementation of step 901 can be similar to that of step 602. The implementation of step 602 will be used as an example to introduce the implementation method of microcrack feature enhancement processing for the image.

[0183] 902. Use the trained defect recognition model to identify defects in the sixth image and output defect recognition information.

[0184] The aforementioned defect identification information indicates whether the photovoltaic module has microcracks or not. Step 902 can be implemented in a similar manner to step 302.

[0185] In this embodiment, the fifth image is first subjected to hidden crack feature enhancement processing to obtain the sixth image; then, the defect recognition model trained is used to perform defect recognition on the sixth image, which can improve the accuracy of defect recognition.

[0186] Figure 10 A flowchart of another photovoltaic module microcrack detection method provided in this application embodiment. Figure 10 The method and process in the middle is Figure 9 Optimization of the methods and processes within the framework. For example... Figure 10 As shown, the method includes:

[0187] 1001. The microcrack detection device performs microcrack feature enhancement processing on the fifth image to obtain the sixth image.

[0188] The implementation method of step 1001 can be the same as that of step 901.

[0189] 1002. The microcrack detection device eliminates the grid lines and boundaries of the sixth image to obtain the seventh image.

[0190] The implementation of step 1002 can be similar to that of step 803.

[0191] 1003. Use the trained defect recognition model to identify defects in the seventh image and output defect recognition information.

[0192] The implementation of step 1003 can be similar to that of step 902.

[0193] Figure 10 Compared with the method flow in Figure 9 The method flow in the embodiment of the present application adds the operation of eliminating the grid lines and the boundaries in the image, so that the grid lines and the boundaries can be avoided to be identified as the hidden crack defects, thereby improving the accuracy of identifying the hidden crack defects in the photovoltaic module.

[0194] Since the foregoing embodiment does not describe the way of performing the hidden crack feature enhancement processing on the image in detail, some possible ways of performing the hidden crack feature enhancement processing on the image are introduced below in combination with the accompanying drawings. Figure 11 A method flow of performing hidden crack feature enhancement processing on an image is provided in the embodiment of the present application. As shown in Figure 11 The method comprises the following steps.

[0195] 1101. The hidden crack detection device obtains F gray scale responses of a third pixel point in the second image in K directions.

[0196] The third pixel point is any pixel point in the second image. The gray scale response of the third pixel point in the first direction is used to determine whether there is a hidden crack in the third pixel point in the first direction, and the first direction is included in the K directions. It should be understood that the gray scale response of the third pixel point in each direction is used to determine whether there is a hidden crack defect in the third pixel point in the direction. K is an integer greater than 1, and F is an integer equal to or greater than K.

[0197] One possible implementation of step 1101 is as follows: obtaining a first gray scale matrix; and obtaining the F gray scale responses of the third pixel point in the K directions according to the first gray scale matrix.

[0198] The directions of the K rays with the center element as the starting point in the first gray scale matrix correspond to the K directions one by one. The first direction corresponds to the third ray in the K rays. The third ray passes through two or more elements in the first gray scale matrix. The gray scale response of the third pixel point in the first direction includes the third ray passing through two or more elements in the first gray scale matrix. The gray scale response of the third pixel point in each direction includes the elements in the first gray scale matrix passed by the ray corresponding to the direction. Figure 12 A schematic diagram of a first gray scale matrix and K rays in the first gray scale matrix is provided in the embodiment of the present application. Figure 12 In the schematic diagram, the center element of the first gray scale matrix is the endpoint of each ray, and each ray passes through two or more elements in the first gray scale matrix. It should be understood that Figure 12This is merely an example, and not to say that K rays can only be 8 rays. The third ray can be... Figure 12 Any ray in it.

[0199] The first grayscale matrix is ​​obtained by dot product of the sliding matrix and the first submatrix. The first element of the sliding matrix is ​​not zero. The second element of the sliding matrix is ​​zero. The direction of the first ray corresponding to the first element is included in the aforementioned K directions. The first ray is a ray in the sliding matrix that starts from the center element and passes through the first element. The direction of the second ray corresponding to the second element is not included in the aforementioned K directions. The second ray is a ray in the sliding matrix that starts from the center element and passes through the second element. A ray corresponding to any element in the sliding matrix is ​​a ray in the sliding matrix that starts from the center element and passes through that element. The sliding matrix can be an M x M matrix, where the directions of the K rays starting from the center element correspond one-to-one with the aforementioned K directions. Each of the K rays passes through two or more elements in the sliding matrix, and the value of each ray passing through an element is 1. The values ​​of elements in the sliding matrix not passed through by the aforementioned K rays are all 0. M is an integer greater than 1, such as 3, 5, 7, etc. For example, the sliding matrix is ​​a 5x5 matrix, with its center element being the element in the 3rd row and 3rd column. In one possible implementation, the angle between any two adjacent rays of the K rays originating from the center element in the sliding matrix is ​​the target angle. For example, if K is 12, the target angle is 24. Or, if K is 18, the target angle is 20. Figure 13 This is a schematic diagram of a sliding matrix and K rays in the sliding matrix provided in an embodiment of this application. Figure 13 In the sliding matrix, some elements have a value of 0, while others have a value of 1. The element in the diamond box is the center element of the sliding matrix, the element in the circle box is an example of the first element, 1301 is an example of the first ray, and the element in the triangle box is an example of the second element. Figure 13 As shown, any one of the K rays has its endpoint at the center element of the sliding matrix and passes through two or more elements of the sliding matrix with a value of 1. In some embodiments, the included angle between any two adjacent rays in the K rays is equal. The center element of the first sub-matrix is ​​the third pixel, and the first sub-matrix is ​​a sub-matrix of the second image. Figure 14 This is a schematic diagram of a second image and a first sub-matrix provided in an embodiment of this application. Figure 14In the figure, the element at the position of the black triangle is an example of the third pixel point, and the group of elements in the smaller dashed rectangular frame constitutes an example of the first sub-matrix; the element at the position of the black circle is another example of the third pixel point, and the group of elements in the larger dashed rectangular frame constitutes another example of the first sub-matrix. The sliding matrix can be understood as a sliding window. The sub-matrix corresponding to a pixel point in the second image can be understood as a matrix covered by the sliding matrix when the center element of the sliding matrix coincides with the pixel point. The hidden crack detection device can multiply the sub-matrix corresponding to each pixel point in the second image with the sliding matrix respectively, so as to obtain the gray matrix corresponding to each pixel point in the second image. The sub-matrix corresponding to a pixel point in the second image is a sub-matrix belonging to the second image with the pixel point as the center element.

[0200] 1102、The hidden crack detection device adjusts the gray value of the third pixel point to a third gray value.

[0201] The third gray value is obtained by using the F gray responses. Figure 11 The method flow of the hidden crack feature enhancement processing of the second image is described by taking the third pixel point in the second image as an example. It should be understood that the hidden crack detection device can adjust the gray value of each pixel point in the second image by using a method flow similar to that in Figure 11 to obtain the third image.

[0202] In a possible implementation, the first gray matrix is an M-row and M-column matrix, and L is an integer greater than 0 and not greater than M. The K rays corresponding to the K directions all pass through L elements in the first gray matrix. The hidden crack detection device can process the third gray value by using the F gray responses of the third pixel point in the K directions. F = L*K. The third pixel point corresponds to L gray responses in each of the K directions.

[0203] In some embodiments, the F gray responses of the third pixel point in the K directions and the third gray value satisfy the following formula:

[0204]

[0205] wherein, represents the third gray value, represents the maximum gray response of the third pixel point in the K directions, represents the average value of the F gray responses, and d is a constant, for example, d is 1. The value of d can be configured as needed. may be the maximum value of the K first sub-grayscale responses in the K directions. The K first sub-grayscale responses correspond to the K directions one by one, and the first sub-grayscale response in each direction is obtained by using the L grayscale responses in the direction. For example, the first sub-grayscale response in a direction is the weighted sum of the F grayscale responses in the direction. For another example, the first sub-grayscale response in a direction is the maximum value of the F grayscale responses in the direction. The L grayscale responses in any direction of the K directions are the L elements in the first grayscale matrix through which the ray corresponding to the direction passes. For example, the L elements in the first grayscale matrix through which the third ray passes in turn are a, b, c, d, and e, and the L grayscale responses in the direction corresponding to the third ray are a, b, c, d, and e; the (L-1) grayscale responses in the direction corresponding to the third ray are a, b, c, and d, and the (L-2) grayscale responses in the direction corresponding to the third ray are a, b, and c; wherein L is 5. may be obtained by using the L*K grayscale responses (i.e., L grayscale responses in each direction) of the third pixel point in the K directions. may be understood as the grayscale response of the linear detector with a length of L to the third pixel point. The crack detection device may also obtain the grayscale response of the linear detector with a length of (L-1) to the third pixel point by using the (L-1)*K grayscale responses (i.e., (L-1) grayscale responses in each direction) of the third pixel point in the K directions.

[0206] The (L-1)*K grayscale responses of the third pixel point in the K directions may satisfy the following formula:

[0207]

[0208] wherein, represents the grayscale response of the linear detector with a length of (L-1) to the third pixel point, represents the maximum value of the K second sub-grayscale responses in the K directions, represents the average value of the (L-1)*K grayscale responses, and d is a constant, for example, d is 1. The K second sub-grayscale responses correspond to the K directions one by one. The second sub-grayscale response in each direction is obtained by using the (L-1) grayscale responses in the direction. It should be understood that the crack detection device may obtain the grayscale responses of two or more linear detectors with different lengths to the third pixel point by using the first grayscale matrix, for example, and

[0209] To increase the robustness of the linear detector, different lengths of linear detectors can be selected in the first gray matrix to calculate, and finally the gray responses of each linear detector to the third pixel point are weighted to obtain the third gray value. In some embodiments, the F gray responses of the third pixel point in the K directions and the third gray value satisfy the following formula:

[0210]

[0211] wherein R represents the third gray value, represents the gray response of the linear detector with a length of L to the third pixel point, P is the gray value of the third pixel point, and n is the scale size of the linear detector (i.e., the number of linear detectors). represents the sum.

[0212] It should be understood that after obtaining the gray responses of each pixel point in the K directions in the second image, the hidden crack detection device can adjust the gray values of each pixel point in a manner similar to adjusting the gray value of the third pixel point.

[0213] In the embodiments of the present application, the gray value of the third pixel point in the second image is adjusted according to the F gray responses of the third pixel point in the K directions, which can effectively enhance the hidden crack features in the second image.

[0214] The foregoing embodiments do not describe the way to train the defect recognition model. The way to train the defect recognition model is introduced as follows. Figure 15 A training method flowchart is provided for the embodiments of the present application. As shown in Figure 15 , the method comprises:

[0215] 1501. The training device inputs the training sample to the initial defect recognition model for processing to obtain a defect recognition result.

[0216] The training device can be the hidden crack detection device or can not be the hidden crack detection device. When the hidden crack detection device and the training device are not the same device, the hidden crack detection device can obtain the defect recognition model trained by the training device from the training device or other devices. The training sample is a sample that has been subjected to hidden crack feature extraction processing and / or hidden crack feature enhancement processing.

[0217] The initial defect identification model can be InceptionV4. The initial defect identification model described above includes a Stem module, an Inception-A module, a Reduction-A module, and an Inception-C module; wherein the second image (i.e. input) first passes through the Stem module, then sequentially passes through the Inception-A module, the Reduction-A module, and the Inception-C module, and then passes through the Softmax classification. Figure 16 A structural diagram of a defect identification model provided by an embodiment of the present application is shown in FIG. 2. As shown in FIG. 2, the defect identification network includes an input layer (Input), a Stem module, an Inception-A module, a Reduction-A module, an Inception-C module, a pooling layer (Average Pooling), a Dropout layer, and a Softmax layer. The pooling layer and the Dropout layer are optional and not necessary. Since InceptionV4 is a commonly used convolutional neural network in the field of deep learning, the functions of the various parts in InceptionV4 will not be described here. Figure 16

[0218] In some embodiments, the training samples can be obtained in the following manner: first, the original samples are converted to obtain intermediate samples; then, each sample in the intermediate samples is subjected to hidden crack feature extraction processing and / or hidden crack feature enhancement processing to obtain the training samples. Considering the rotation feature invariance of the convolutional neural network, the original samples (i.e. the collected EL images) can be converted to obtain more intermediate samples. This can reduce the number of EL images required for training the defect identification model. In actual applications, after obtaining the training samples, each training sample can be labeled, i.e. adding a label to each training sample indicating whether the training sample has a hidden crack defect. After each training sample is labeled, the defect identification model can be trained using the method flow in FIG. 1. Figure 14

[0219] 1502. The training device determines the loss corresponding to the training sample according to the defect identification result and the standard result.

[0220] The standard result is the processing result expected to be obtained by processing the training sample using the initial defect identification model, i.e. the ideal result. The standard result can be understood as a label added to each training sample in advance.

[0221] 1503. The training device updates the parameters of the initial defect identification network by an optimization algorithm using the loss corresponding to the training sample to obtain the defect identification model.

[0222] ​​The training device can update parameters of the defect recognition model by using the loss corresponding to the training sample and by using a gradient descent method.

[0223] In the embodiments of the present application, the initial defect recognition model is trained by using samples subjected to the hidden crack feature extraction processing and / or the hidden crack feature enhancement processing, so as to obtain the defect recognition model. A smaller number of training samples can be used to train the defect recognition model with higher precision.

[0224] The foregoing embodiments introduce the method for detecting the hidden crack of the photovoltaic module. The structure of the hidden crack detection device for implementing the method is introduced below. Figure 17 FIG. 1 shows a structure diagram of a hidden crack detection device according to an embodiment of the present application. As shown in FIG. 1, the hidden crack detection device can include: Figure 17

[0225] The feature extraction processing unit 1701 is configured to perform hidden crack feature extraction processing on the first image to obtain a second image. The first image is an electroluminescence (EL) image obtained by detecting the photovoltaic module. The hidden crack feature extraction processing includes adjusting a gray value of a first pixel point to a first gray value and adjusting a gray value of a second pixel point to a second gray value. The first pixel point and the second pixel point are both included in the first image. A first eigenvalue of a first Hessian matrix corresponding to the first pixel point satisfies a target condition. A second eigenvalue of a second Hessian matrix corresponding to the second pixel point does not satisfy the target condition. The first gray value is obtained by using the first eigenvalue. The second gray value is different from the first gray value.

[0226] The defect recognition unit 1702 is configured to perform defect recognition on the second image by using the trained defect recognition model to obtain a defect recognition result. The defect recognition result indicates whether the photovoltaic module has a hidden crack or not.

[0227] The output unit 1703 is configured to output the defect recognition result.

[0228] In one possible implementation, the first eigenvalue satisfying the target condition includes that a third eigenvalue included in the first eigenvalue is less than or equal to 0. The third eigenvalue is an eigenvalue with a larger absolute value in two eigenvalues included in the first eigenvalue. The second eigenvalue not satisfying the target condition includes that a fourth eigenvalue included in the second eigenvalue is greater than 0. The fourth eigenvalue is an eigenvalue with a larger absolute value in two eigenvalues included in the second eigenvalue.

[0229] ​In a possible implementation, the feature extraction processing unit 1701 is further configured to construct the first Hessian matrix corresponding to the first pixel point and construct the second Hessian matrix corresponding to the second pixel point; and process the first Hessian matrix to obtain the first eigenvalue and process the second Hessian matrix to obtain the second eigenvalue.

[0230] In a possible implementation, the hidden crack detection apparatus further includes a feature enhancement processing unit 1704 configured to perform hidden crack feature enhancement processing on the second image to obtain a third image; the hidden crack feature enhancement processing includes obtaining F gray scale responses of a third pixel point in the second image in K directions, and adjusting a gray scale value of the third pixel point to a third gray scale value; the third gray scale value is obtained by using the F gray scale responses, and a gray scale response of the third pixel point in a first direction is used to determine whether there is a hidden crack in the first direction of the third pixel point, the first direction is included in the K directions, K is an integer greater than 1, and F is an integer equal to or greater than K.

[0231] The defect identification unit 1702 is specifically configured to perform defect identification on the third image by using the trained defect identification model to obtain the defect identification result.

[0232] In a possible implementation, the feature enhancement processing unit 1704 is specifically configured to obtain a first gray scale matrix; the first gray scale matrix is obtained by point multiplication of a sliding matrix and a first sub-matrix, a first element in the sliding matrix is not zero, a second element in the sliding matrix is zero, a direction of a first ray corresponding to the first element is included in the K directions, the first ray is a ray in the sliding matrix that starts from a center element and passes through the first element, a direction of a second ray corresponding to the second element is not included in the K directions, the second ray is a ray in the sliding matrix that starts from the center element and passes through the second element, and a center element of the first sub-matrix is the third pixel point and the first sub-matrix is a sub-matrix of the second image; according to the first gray scale matrix, the F gray scale responses of the third pixel point in the K directions are obtained; K rays in the first gray scale matrix that start from the center element correspond to the K directions one by one, the first direction corresponds to a third ray in the K rays, the third ray passes through two or more elements in the first gray scale matrix, and the gray scale response of the third pixel point in the first direction includes the third ray passing through the two or more elements in the first gray scale matrix.

[0233] In one possible implementation, the first grayscale matrix is ​​an M-row, M-column matrix, the third ray passes through L elements of the first grayscale matrix, where M is an odd number greater than 1 and L is an integer greater than 0 and not greater than M; the feature enhancement processing unit 1704 is further configured to process the F grayscale responses of the third pixel in the K directions to obtain the third grayscale value.

[0234] In one possible implementation, the feature enhancement processing unit 1704 is also used to normalize the gray values ​​of each pixel in the third image to obtain the fourth image.

[0235] The defect recognition unit 1702 is specifically used to perform defect recognition on the fourth image using the defect recognition model obtained through training, and to obtain the defect recognition result.

[0236] In one possible implementation, the above-mentioned microcrack detection device further includes: an interference cancellation unit 1705, used to eliminate the grid lines and the boundaries of the third image to obtain a fourth image; and a defect recognition unit, specifically used to use the trained defect recognition model to perform defect recognition on the fourth image to obtain the defect recognition result.

[0237] Figure 18 This is a schematic diagram of a microcrack detection device provided in an embodiment of this application. Figure 18 As shown, the microcrack detection device may include:

[0238] The microcrack enhancement unit 1801 is used to perform microcrack feature enhancement processing on the fifth image to obtain the sixth image; the fifth image is an electroluminescent EL image obtained by detecting a photovoltaic module; the microcrack feature enhancement processing includes: obtaining F grayscale responses of the fourth pixel in the fifth image in K directions, and adjusting the grayscale value of the fourth pixel to the fourth grayscale value; the fourth grayscale value is obtained using the F grayscale responses; the grayscale response of the fourth pixel in the second direction is used to determine whether there is a microcrack in the fourth pixel in the second direction; the second direction is included in the K directions; K is an integer greater than 1; and F is an integer equal to or greater than K.

[0239] The defect processing unit 1802 is used to perform defect recognition on the sixth image using the trained defect recognition model to obtain defect recognition information; the defect recognition information indicates whether the photovoltaic module has hidden cracks or does not have hidden cracks.

[0240] Output unit 1803 is used to output the above-mentioned defect identification information.

[0241] In a possible implementation, the hidden crack enhancement unit 1801 is specifically configured to: obtain a second gray matrix; the second gray matrix is obtained by point multiplication of a sliding matrix and a second sub-matrix, a first element in the sliding matrix is not zero, a second element in the sliding matrix is zero, a first ray corresponding to the first element is in the K directions, the first ray is a ray in the sliding matrix starting from a center element and passing through the first element, a second ray corresponding to the second element is not in the K directions, the second ray is a ray in the sliding matrix starting from the center element and passing through the second element, the center element of the second sub-matrix is the fourth pixel point, and the second sub-matrix is a sub-matrix of the fifth image; obtain the F gray responses of the fourth pixel point in the K directions according to the second gray matrix; K rays starting from the center element in the second gray matrix correspond to the K directions one by one, the second direction corresponds to a fourth ray in the K rays, the fourth ray passes through two or more elements in the second gray matrix, and the gray response of the fourth pixel point in the second direction includes the fourth ray passing through the two or more elements in the second gray matrix.

[0242] In a possible implementation, the second gray matrix is an M-row-by-M-column matrix, the fourth ray passes through L elements in the second gray matrix, M is an odd number greater than 1, and L is an integer greater than 0 and not greater than M; the hidden crack enhancement unit 1801 is further configured to obtain the fourth gray value by processing the F gray responses of the fourth pixel point in the K directions.

[0243] In a possible implementation, the hidden crack detection apparatus further includes an interference processing unit 1804 configured to eliminate a grid line in the sixth image and a boundary of the sixth image to obtain a seventh image.

[0244] The defect processing unit 1802 is specifically configured to perform defect identification on the seventh image by using the trained defect identification model to obtain the defect identification information.

[0245] Figure 19 is a block diagram of part of the structure of an electronic device provided by an embodiment of the present application. As shown in Figure 19 the electronic device 1900 can include a processor 1901, a memory 1902, an input device 1903, an output device 1904, and a bus 1905. The processor 1901, the memory 1902, the input device 1903, and the output device 1904 can be communicatively connected to each other through the bus 1905. The bus 1905 can be divided into an address bus, a data bus, a control bus, and the like. For ease of representation,Figure 19 Only one bus is shown in FIG. 19 for simplicity, but there can be more than one bus, and various buses can be used to implement the buses illustrated in FIG. 19.

[0246] The processor 1901 can be a general purpose central processing unit (CPU), a microprocessor, a graphics processing unit (GPU), an application specific integrated circuit (ASIC), or one or more integrated circuits, for executing programs to implement the techniques described in the embodiments of the present application. The processor 1901 can implement the functions of the feature extraction processing unit 1701, the defect identification unit 1702, the feature enhancement processing unit 1704, and the interference elimination unit 1705 in FIG. 17, and can also implement the functions of the crack enhancement unit 1801, the defect processing unit 1802, and the interference processing unit 1804 in FIG. 18. Figure 17 Figure 18

[0247] The memory 1902 can be a read only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1902 can store an operating system, and other application programs. The program codes required for the modules and components included in the electronic device to perform the functions provided by the software or firmware, or the program codes required for implementing the above-described methods provided by the method embodiments of the present application are stored in the memory 1902, and the processor 1901 reads the codes in the memory 1902 to perform the operations required for the modules and components included in the electronic device, or to perform the methods provided by the embodiments of the present application.

[0248] The input device 1903 is used to input the image to be processed by the electronic device, such as the first image.

[0249] The output device 1904 is used to output the defect identification result obtained by the electronic device. The output device 1904 can implement the function of the output unit 1703 in FIG. 17, and can also implement the function of the output unit 1803 in FIG. 18. Figure 17 Figure 18

[0250] The bus 1905 can include a path for transmitting information between the various components (e.g., the processor 1901, the memory 1902, the input device 1903, and the output device 1904) of the electronic device.

[0251] It should be noted that although the above-described embodiments of the present application are described in the context of a mobile terminal, the present application is not limited to this. The present application can be applied to any electronic device having a display screen, such as a personal computer, a personal digital assistant (PDA), a digital camera, a digital camcorder, a digital music player, a mobile phone, a smart phone, a tablet computer, a wearable device, and a smart watch. Figure 19 ​​​​The electronic device 1900 shown only shows the processor 1901, the memory 1902, the input device 1903, the output device 1904, and the bus 1905, but in the specific implementation process, those skilled in the art should understand that the electronic device 1900 also contains other devices necessary for normal operation. At the same time, according to the specific needs, those skilled in the art should understand that the electronic device 1900 can also contain hardware devices that realize other additional functions. In addition, those skilled in the art should understand that the electronic device 1900 can also only contain devices necessary for the embodiments of the present application, and does not necessarily contain all the devices shown in the electronic device 1900. Figure 19 The electronic device 1900 shown only shows the processor 1901, the memory 1902, the input device 1903, the output device 1904, and the bus 1905, but in the specific implementation process, those skilled in the art should understand that the electronic device 1900 also contains other devices necessary for normal operation. At the same time, according to the specific needs, those skilled in the art should understand that the electronic device 1900 can also contain hardware devices that realize other additional functions. In addition, those skilled in the art should understand that the electronic device 1900 can also only contain devices necessary for the embodiments of the present application, and does not necessarily contain all the devices shown in the electronic device 1900.

[0252] Figure 20 Another structure schematic diagram of an electronic device 200 provided by the embodiments of the present application is shown. As shown in the figure, Figure 20 Figure 20 The electronic device shown includes a logic circuit 2001 and an interface 2002. The logic circuit 2001 can realize the functions of the feature extraction processing unit 1701, the defect identification unit 1702, the feature enhancement processing unit 1704, and the interference elimination unit 1705 in the Figure 17 . Alternatively, the logic circuit 2001 can realize the functions of the hidden crack enhancement unit 1801, the defect processing unit 1802, and the interference processing unit 1804 in the Figure 18 . The interface 2002 can realize the function of the output unit 1703 in the Figure 17 . Alternatively, the interface 2002 can realize the function of the output unit 1803 in the Figure 18 . Wherein, the logic circuit 2001 can be a chip, a processing circuit, an integrated circuit, or a system on chip (SoC) chip, etc., and the interface 2002 can be a communication interface, an input and output interface, etc. In the embodiments of the present application, the logic circuit and the interface can also be coupled to each other. The specific connection mode of the logic circuit and the interface is not limited in the embodiments of the present application.

[0253] The present application also provides a computer readable storage medium, which stores computer code, when the computer code runs on the computer, makes the computer execute the method of the above-mentioned embodiments.

[0254] The present application also provides a computer program product, which includes computer code or computer program, when the computer code or computer program runs on the computer, makes the authentication and authorization method in the above-mentioned embodiments be executed.

[0255] ​The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the above claims.

Claims

1. A method for detecting a hidden crack in a photovoltaic module, the method comprising: The method comprises: performing a hidden crack feature extraction process on the first image to obtain a second image; the first image is an electroluminescence (EL) image obtained by detecting a photovoltaic module, and the hidden crack feature extraction process comprises: adjusting a gray value of a first pixel point to a first gray value, and adjusting a gray value of a second pixel point to a second gray value; the first pixel point and the second pixel point are both included in the first image, a first eigenvalue of a first Hessian matrix corresponding to the first pixel point satisfies a target condition, a second eigenvalue of a second Hessian matrix corresponding to the second pixel point does not satisfy the target condition, the first gray value is obtained by using the first eigenvalue, and the second gray value is different from the first gray value; performing defect recognition on the second image by using a trained defect recognition model to output a defect recognition result; the defect recognition result indicates whether the photovoltaic module has hidden cracks or not; after performing the hidden crack feature extraction process on the first image to obtain the second image, the method further comprises: performing a hidden crack feature enhancement process on the second image to obtain a third image; the hidden crack feature enhancement process comprises: obtaining F gray responses of a third pixel point in the second image in K directions, and adjusting a gray value of the third pixel point to a third gray value; the third gray value is obtained by using the F gray responses, a gray response of the third pixel point in a first direction is used to determine whether the third pixel point has hidden cracks in the first direction, the first direction is included in the K directions, K is an integer greater than 1, and F is an integer equal to or greater than K; the defect recognition on the second image by using the trained defect recognition model to output the defect recognition result comprises: performing defect recognition on the third image by using the trained defect recognition model to output the defect recognition result.

2. The method of claim 1, wherein, the first eigenvalue satisfying the target condition comprises: a third eigenvalue included in the first eigenvalue is less than or equal to 0, the third eigenvalue is an eigenvalue with a larger absolute value among two eigenvalues included in the first eigenvalue, and the second eigenvalue not satisfying the target condition comprises: a fourth eigenvalue included in the second eigenvalue is greater than 0, the fourth eigenvalue is an eigenvalue with a larger absolute value among two eigenvalues included in the second eigenvalue.

3. The method of claim 1, wherein, the obtaining of the F gray responses of the third pixel point in the second image in the K directions comprises: obtaining a first gray matrix; the first gray matrix is obtained by point multiplication of a sliding matrix and a first sub-matrix, a first element in the sliding matrix is not zero, a second element in the sliding matrix is zero, a first ray corresponding to the first element contains a direction in the K directions, the first ray is a ray in the sliding matrix starting from a center element and passing through the first element, a second ray corresponding to the second element does not contain a direction in the K directions, the second ray is a ray in the sliding matrix starting from the center element and passing through the second element, a center element of the first sub-matrix is the third pixel point, and the first sub-matrix is a sub-matrix of the second image; obtaining the F gray responses of the third pixel point in the K directions according to the first gray matrix; K rays starting from a center element in the first gray matrix correspond to the K directions one by one, the first direction corresponds to a third ray in the K rays, the third ray passes through two or more elements in the first gray matrix, and the gray response of the third pixel point in the first direction includes the third ray passing through the two or more elements in the first gray matrix.

4. The method of claim 1, wherein, After performing the hidden crack feature enhancement processing on the second image to obtain a third image, the method further comprises: eliminating the grid lines in the third image and the boundary of the third image to obtain a fourth image; the defect recognition result output by the defect recognition model trained and outputting the defect recognition result include: performing defect recognition on the fourth image by using the defect recognition model trained to obtain the defect recognition result.

5. A crack detection apparatus characterized by comprising: comprise: a feature extraction processing unit configured to perform hidden crack feature extraction processing on a first image to obtain a second image; the first image is an electroluminescence (EL) image obtained by detecting a photovoltaic assembly, and the hidden crack feature extraction processing comprises: adjusting a gray value of a first pixel point to a first gray value, and adjusting a gray value of a second pixel point to a second gray value; the first pixel point and the second pixel point are both included in the first image, a first eigenvalue of a first Hessian matrix corresponding to the first pixel point satisfies a target condition, a second eigenvalue of a second Hessian matrix corresponding to the second pixel point does not satisfy the target condition, the first gray value is obtained by using the first eigenvalue, and the second gray value is different from the first gray value; a defect recognition unit configured to perform defect recognition on the second image by using a defect recognition model trained to obtain a defect recognition result; the defect recognition result indicates whether the photovoltaic assembly has a hidden crack or not; an output unit configured to output the defect recognition result; the hidden crack detection device further comprises: The feature enhancement processing unit is configured to perform hidden crack feature enhancement processing on the second image to obtain a third image. The hidden crack feature enhancement processing includes: obtaining F gray scale responses of a third pixel point in the second image in K directions, and adjusting a gray scale value of the third pixel point to a third gray scale value. The third gray scale value is obtained by using the F gray scale responses. A gray scale response of the third pixel point in a first direction is used to determine whether there is a hidden crack in the first direction. The first direction is included in the K directions. K is an integer greater than 1. F is an integer equal to or greater than K. The defect identification unit is specifically configured to perform defect identification on the third image by using the trained defect identification model to obtain the defect identification result.

6. The crack detection apparatus according to claim 5, wherein The first feature value satisfying the target condition includes that a third feature value included in the first feature value is less than or equal to 0. The third feature value is a feature value with a larger absolute value in two feature values included in the first feature value. The second feature value not satisfying the target condition includes that a fourth feature value included in the second feature value is greater than 0. The fourth feature value is a feature value with a larger absolute value in two feature values included in the second feature value.

7. The hidden crack detection device according to claim 5, characterized in that, The feature enhancement processing unit is specifically configured to obtain a first gray scale matrix. The first gray scale matrix is obtained by point multiplication of a sliding matrix and a first sub-matrix. A first element in the sliding matrix is not zero. A second element in the sliding matrix is zero. A direction of a first ray corresponding to the first element is included in the K directions. The first ray is a ray in the sliding matrix starting from a center element and passing through the first element. A direction of a second ray corresponding to the second element is not included in the K directions. The second ray is a ray in the sliding matrix starting from the center element and passing through the second element. A center element of the first sub-matrix is the third pixel point, and the first sub-matrix is a sub-matrix of the second image. According to the first gray scale matrix, the F gray scale responses of the third pixel point in the K directions are obtained. Directions of K rays starting from a center element in the first gray scale matrix correspond to the K directions one by one. The first direction corresponds to a third ray in the K rays. The third ray passes through two or more elements in the first gray scale matrix. The gray scale response of the third pixel point in the first direction includes the third ray passing through the two or more elements in the first gray scale matrix.

8. The crack detection apparatus according to claim 5, wherein The hidden crack detection device further includes: An interference elimination unit is configured to eliminate a grid line in the third image and a boundary of the third image to obtain a fourth image. The defect identification unit is specifically configured to perform defect identification on the fourth image by using the trained defect identification model to obtain the defect identification result.

9. A photovoltaic module crack detection system, comprising: The hidden crack detection device further includes: The photovoltaic power station management device, the data collector, the control device, the plurality of inverters, the plurality of photovoltaic strings, the hidden crack detection device, and the image acquisition device; The image acquisition device is configured to move according to a control instruction from the control device; Send the pose information to the control device; The pose information is used to indicate the pose of the image acquisition device; The control device is configured to send a back-feeding current request to the data collector; The back-feeding current request contains the identification of a target inverter; The target inverter is an inverter in the plurality of inverters that is coupled to a target photovoltaic string, and the target photovoltaic string belongs to the photovoltaic string photographed by the image acquisition device at the pose. The data collector is configured to send the back-feeding current request to the photovoltaic power station management device; The photovoltaic power station management device is configured to send a back-feeding current instruction to the target inverter; The target inverter is configured to apply a back-feeding current to the target photovoltaic string after receiving the back-feeding current instruction. The image acquisition device is configured to acquire an electroluminescence (EL) image of the target photovoltaic string to obtain a first image. The hidden crack detection device is configured to obtain the first image and perform hidden crack detection using the first image to obtain a defect identification result. The defect identification result indicates whether a target photovoltaic component in the target photovoltaic string has a hidden crack. The hidden crack detection device is specifically configured to perform hidden crack feature extraction processing on the first image to obtain a second image. The hidden crack feature extraction processing includes adjusting a gray value of a first pixel point to a first gray value and adjusting a gray value of a second pixel point to a second gray value. The first pixel point and the second pixel point are both included in the first image. A first eigenvalue of a first Hessian matrix corresponding to the first pixel point satisfies a target condition, a second eigenvalue of a second Hessian matrix corresponding to the second pixel point does not satisfy the target condition, the first gray value is obtained using the first eigenvalue, and the second gray value is different from the first gray value. The hidden crack feature enhancement processing includes obtaining F gray responses of a third pixel point in the second image in K directions and adjusting a gray value of the third pixel point to a third gray value. The third gray value is obtained using the F gray responses. A gray response of the third pixel point in a first direction is used to determine whether the third pixel point has a hidden crack in the first direction. The first direction is included in the K directions. K is an integer greater than 1, and F is an integer equal to or greater than K. A trained defect identification model is used to perform defect identification on the third image to output the defect identification result.

10. The photovoltaic module crack detection system of claim 9, wherein, The image acquisition device is a drone carrying an infrared camera, and the control device is a remote controller used to control the drone. The control device is further configured to determine the target photovoltaic string from the photovoltaic string photographed by the image acquisition device according to the pose information.

Images