A method and system for evaluating the anti-reflection performance of mobile phone tempered glass film

Abstract

The present invention discloses a method and system for evaluating the anti-reflection performance of a mobile phone tempered glass film, comprising: S1: irradiating the surface of the tempered glass film with a multi-angle light source array, obtaining image sequences under different light sources through high dynamic range imaging technology and calculating the reflected light intensity feature image sequence; S2: performing non-downsampled contour wave decomposition on the reflected light intensity feature image sequence, and obtaining a comprehensive reflection feature map through an adaptive weight fusion algorithm; S3: constructing a deep learning feature extraction network, extracting features from the reflected light intensity feature image sequence and the comprehensive reflection feature map, and performing feature fusion; S4: extracting multi-scale features using a feature pyramid network, and performing feature aggregation; S5: mapping the aggregated multi-scale features based on a fully connected network, and obtaining an anti-reflection performance evaluation score through an activation function. The present invention can realize the automated and intelligent evaluation of the anti-reflection performance of the tempered glass film.

Description

Technical Field

[0001] The present invention relates to the technical field of anti-reflection performance evaluation, and in particular to a method and system for evaluating the anti-reflection performance of a mobile phone tempered glass film. Background Art

[0002] With the popularity of smart phones, tempered glass films have become an important accessory for mobile phone screen protection. Among them, anti-reflection performance is one of the key indicators for evaluating the quality of tempered glass films, which directly affects the user's display perception and usage experience. At present, the anti-reflection performance evaluation method of mobile phone tempered glass films mainly has the following problems:

[0003] Traditional evaluation methods mainly rely on manual visual inspection or single-angle reflectivity measurement, which has the disadvantages of strong subjectivity, poor repeatability, and incomplete evaluation results. Existing instrument measurement methods mostly use reflectivity measurement at a single incident angle, such as the spectrophotometer method, which cannot fully characterize the anti-reflection properties under different viewing angles. This method ignores the complex influence of multi-angle light sources in the actual use environment, and there is a large difference between the evaluation results and the actual experience. Existing feature extraction methods are mostly based on traditional image processing algorithms, such as grayscale statistical features, texture features, etc., which are difficult to effectively characterize the intrinsic relationship between the surface microstructure of the glass film and the optical properties. Summary of the invention

[0004] In view of this, the present invention provides a method and system for evaluating the anti-reflection performance of a mobile phone tempered glass film, aiming to achieve automatic and intelligent evaluation of the anti-reflection performance of the tempered glass film by combining a multi-angle light source array acquisition system, non-subsampled contourlet decomposition feature extraction and a deep learning network.

[0005] To achieve the above object, the present invention provides a method for evaluating the anti-reflection performance of a mobile phone tempered glass film, comprising the following steps:

[0006] S1: Use a multi-angle light source array to illuminate the surface of the tempered glass film, obtain image sequences under different light sources through high dynamic range imaging technology, and calculate the reflected light intensity characteristic image sequence;

[0007] S2: Perform non-subsampled contourlet decomposition on each image in the reflected light intensity feature image sequence to obtain non-subsampled contourlet coefficients, calculate local energy features and directional consistency of the non-subsampled contourlet coefficients; calculate adaptive fusion weights based on the local energy features and directional consistency; fuse the adaptive fusion weights with the non-subsampled contourlet coefficients to obtain fused non-subsampled contourlet coefficients; perform inverse transformation on the fused non-subsampled contourlet coefficients to obtain a comprehensive reflection feature map;

[0008] S3: Build a deep learning feature extraction network, extract features from the reflected light intensity feature image sequence and the comprehensive reflection feature map, and perform feature fusion to obtain the fused feature map;

[0009] S4: Use the feature pyramid network to extract the multi-scale features of the fused feature map, and perform feature aggregation to obtain the aggregated multi-scale features;

[0010] S5: Map the aggregated multi-scale features based on the fully connected network, and obtain the anti-reflection performance evaluation score through the activation function.

[0011] As a further improvement method of the present invention:

[0012] Optionally, in step S1, a multi-angle light source array is used to illuminate the surface of the tempered glass film, an image sequence under different light sources is acquired by high dynamic range imaging technology, and a characteristic image sequence of reflected light intensity is calculated, including:

[0013] S11: Construct a circular array of 8 LED light sources, with the light sources evenly distributed along the circumference. is the light source number, and the The azimuth of the light source and The incident angle of the light source ;

[0014] S12: Acquire image sequences under different light sources ,in For the Images acquired under illumination of light sources, using high dynamic range imaging technology, with three exposure times collected at each light source position The following image is synthesized into a high dynamic range image:

[0015] ;

[0016] in, For the The image obtained under the illumination of a light source is at the pixel point The pixel value at ; are the horizontal and vertical coordinates of the image; For the Under the light source Exposure time The collected image is at pixel point The pixel value at ; is the weight function, specifically:

[0017] ;

[0018] in, and Respectively Under the light source Exposure time The minimum and maximum pixel values ​​of all pixels of the collected image;

[0019] S13: Constructing the angular response feature matrix And calculate the reflected light intensity characteristic image sequence .

[0020] Optionally, the step S13 includes:

[0021] S131: Calculate the The angle response characteristic matrix at the pixel point under the illumination of the light source The response value at :

[0022] ;

[0023] in, Pixel To The distance between the projection centers of the light sources; is the focal length of the imaging system;

[0024] S132: Calculate the statistical characteristics of the reflected light intensity of each pixel:

[0025] ;

[0026] in, is the average reflection intensity feature In pixels The pixel value at ; is the reflection intensity standard deviation characteristic In pixels The pixel value at ; is the reflection intensity dynamic range characteristic In pixels The pixel value at ; and They are , and The maximum and minimum values ​​in the image sequence of reflected light intensity characteristics .

[0027] Optionally, in the step S2, each image in the reflected light intensity characteristic image sequence is subjected to non-subsampled contourlet decomposition to obtain non-subsampled contourlet coefficients, and local energy features and directional consistency of the non-subsampled contourlet coefficients are calculated; based on the local energy features and directional consistency, adaptive fusion weights are calculated; the adaptive fusion weights are fused with the non-subsampled contourlet coefficients to obtain fused non-subsampled contourlet coefficients; and the fused non-subsampled contourlet coefficients are inversely transformed to obtain a comprehensive reflection characteristic map, including:

[0028] S21: Reflected light intensity feature image sequence Each image in Perform a non-subsampled Contourlet transform decomposition:

[0029] ;

[0030] in, For the The feature image is Tier Pixels in each direction The non-subsampled contourlet coefficients at ; is the non-subsampled contourlet transform decomposition function; for In pixels The pixel value at ; is the number of decomposition layers; For the The number of directions of the layer, ;

[0031] S22: Calculate the local energy characteristics and directional consistency of each non-subsampled contourlet coefficient:

[0032] ;

[0033] in, and They are The local energy characteristics and directional consistency of is the Gaussian weight function, , is a natural constant, and are the horizontal and vertical indices of the Gaussian kernel weight function, is the local window radius; For the The feature image is Tier Pixels in each direction The non-subsampled contourlet coefficients at ;

[0034] S23: Based on the local energy characteristics and directional consistency, the adaptive fusion weights are calculated; the adaptive fusion weights are fused with the non-subsampled contour wave coefficients to obtain the fused non-subsampled contour wave coefficients; the fused non-subsampled contour wave coefficients are inversely transformed to obtain a comprehensive reflection feature map.

[0035] Optionally, the step S23 includes:

[0036] S231: Calculate adaptive fusion weights based on local energy features and directional consistency:

[0037] ;

[0038] in, for Adaptive fusion weights of and They are The local energy characteristics and directional consistency of For the The feature image is Tier Pixels in each direction The non-subsampled contourlet coefficients at ;

[0039] S232: Fusing the adaptive fusion weights with the non-subsampled contourlet coefficients to obtain fused non-subsampled contourlet coefficients:

[0040] ;

[0041] in, After fusion Tier Pixels in each direction The non-subsampled contourlet coefficients at ;

[0042] S233: Perform an inverse transformation on the fused non-subsampled contourlet coefficients to obtain a comprehensive reflection feature map, specifically:

[0043] ;

[0044] in, Comprehensive reflection feature map In pixels The pixel value at ; is the non-subsampled contourlet inverse transform function.

[0045] Optionally, in the step S3, a deep learning feature extraction network is constructed to extract features from the reflected light intensity feature image sequence and the comprehensive reflection feature map, and feature fusion is performed to obtain a fused feature map, including:

[0046] S31: Construct two convolutional neural networks with the same structure to analyze the reflected light intensity feature image sequence respectively. and comprehensive reflection characteristic map Perform feature extraction and generate the first convolution feature map respectively And the second convolutional feature map ; Each convolutional neural network consists of 5 convolutional layers, using Convolution kernel is used for feature extraction;

[0047] S32: Perform feature fusion:

[0048] ;

[0049] in, is the fused feature map In pixels The eigenvalue at ; and They are the first convolutional feature maps respectively. And the second convolutional feature map In pixels The eigenvalue at ; is the adaptive weight coefficient at the pixel point The coefficient value at is:

[0050] ;

[0051] in, and is a trainable weight matrix.

[0052] Optionally, in step S4, a feature pyramid network is used to extract multi-scale features of the fused feature graph, and feature aggregation is performed to obtain aggregated multi-scale features, including:

[0053] S41: Construct feature pyramid:

[0054] ;

[0055] in, For the pyramid levels; For the Layer feature pyramid at pixel point The eigenvalue at ; and They are upsampling and downsampling operations respectively; For the Convolution operation of the layer;

[0056] S42: Perform feature aggregation:

[0057] ;

[0058] in, The multi-scale features after aggregation are at the pixel point The eigenvalue at ; For the The weight coefficient of the layer feature.

[0059] Optionally, in step S5, mapping the aggregated multi-scale features based on a fully connected network and obtaining an anti-reflection performance evaluation score through an activation function includes:

[0060] S51: Constructing the fully connected layer:

[0061] ;

[0062] in, is the feature vector; It is a fully connected layer operation; For the flattening operation;

[0063] S52: Calculate the evaluation score:

[0064] ;

[0065] in, is the anti-reflection performance evaluation score, ranging from ; is the activation function; and The weight matrix and bias term are evaluated for anti-reflection performance, respectively.

[0066] The present invention also discloses a mobile phone tempered glass film anti-reflection performance evaluation system, comprising:

[0067] High dynamic range imaging module: uses a multi-angle light source array to illuminate the surface of the tempered glass film, obtains image sequences under different light sources through high dynamic range imaging technology, and calculates the reflected light intensity characteristic image sequence;

[0068] Non-subsampled contourlet decomposition module: perform non-subsampled contourlet decomposition on each image in the reflected light intensity feature image sequence to obtain non-subsampled contourlet coefficients, calculate the local energy characteristics and directional consistency of the non-subsampled contourlet coefficients; calculate the adaptive fusion weights based on the local energy characteristics and directional consistency; fuse the adaptive fusion weights with the non-subsampled contourlet coefficients to obtain the fused non-subsampled contourlet coefficients; perform inverse transformation on the fused non-subsampled contourlet coefficients to obtain a comprehensive reflection feature map;

[0069] Feature extraction module: build a deep learning feature extraction network, extract features from the reflected light intensity feature image sequence and the comprehensive reflection feature map, and perform feature fusion to obtain the fused feature map;

[0070] Feature aggregation module: The feature pyramid network is used to extract the multi-scale features of the fused feature map, and then feature aggregation is performed to obtain the aggregated multi-scale features;

[0071] Anti-reflection performance evaluation module: Maps the aggregated multi-scale features based on a fully connected network, and obtains the anti-reflection performance evaluation score through the activation function.

[0072] Compared with the prior art, the present invention has at least the following beneficial effects:

[0073] The present invention uses an annular array of 8 LED light sources and high dynamic range imaging technology, combined with a non-subsampled contourlet decomposition algorithm, to fully capture the reflection characteristics under different incident angles. Through multi-exposure image synthesis and adaptive weight fusion, the problem of overexposure in strong reflection areas and information loss in weak reflection areas is effectively solved, significantly improving the accuracy and completeness of feature extraction.

[0074] The present invention uses a deep learning architecture that combines a dual convolutional neural network and a feature pyramid network to achieve multi-scale fusion of reflection features through an adaptive weight mechanism. This architecture can not only process local detail features and global structural features at the same time, but also adaptively adjust the importance weights of different features, making the evaluation results more comprehensive and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 The present invention is a flowchart of a method for evaluating the anti-reflection performance of a tempered glass film for a mobile phone according to an embodiment of the present invention. DETAILED DESCRIPTION

[0076] The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0077] Example 1: A method for evaluating the anti-reflection performance of a mobile phone tempered glass film, such as Figure 1 As shown, the following steps are included:

[0078] S1: Use a multi-angle light source array to illuminate the surface of the tempered glass film, obtain image sequences under different light sources through high dynamic range imaging technology, and calculate the reflected light intensity characteristic image sequence:

[0079] S11: Construct a circular array of 8 LED light sources, with the light sources evenly distributed along the circumference. is the light source number, The azimuth of the light source and The incident angle of the light source They are:

[0080] ;

[0081] in, is the ratio of pi;

[0082] S12: Acquire image sequences under different light sources ,in For the Images acquired under illumination of light sources, using high dynamic range imaging technology, with three exposure times collected at each light source position The following image is synthesized into a high dynamic range image:

[0083] ;

[0084] in, For the The image obtained under the illumination of a light source is at the pixel point The pixel value at ; are the horizontal and vertical coordinates of the image; For the Under the light source Exposure time The collected image is at pixel point The pixel value at ; is the weight function, specifically:

[0085] ;

[0086] in, and Respectively Under the light source Exposure time The minimum and maximum pixel values ​​of all pixels of the collected image;

[0087] S13: Constructing the angular response feature matrix And calculate the reflected light intensity characteristic image sequence :

[0088] S131: Calculate the The angle response characteristic matrix at the pixel point under the illumination of the light source The response value at :

[0089] ;

[0090] in, Pixel To The distance between the projection centers of the light sources; is the focal length of the imaging system;

[0091] S132: Calculate the statistical characteristics of the reflected light intensity of each pixel:

[0092] ;

[0093] in, is the average reflection intensity feature In pixels The pixel value at ; is the reflection intensity standard deviation characteristic In pixels The pixel value at ; is the reflection intensity dynamic range characteristic In pixels The pixel value at ; and They are , and The maximum and minimum values ​​in the image sequence of reflected light intensity characteristics .

[0094] This step introduces high dynamic range imaging technology, which collects three images with different exposure times for each light source position and synthesizes them through an adaptive weight function. This method effectively solves the problem of overexposure of strong reflection areas and loss of information in weak reflection areas in traditional single exposure imaging, and significantly improves the quality and reliability of image data. In particular, in the design of the weight function, the dynamic range characteristics of the pixel value are taken into account, and targeted optimization processing is performed on different brightness areas.

[0095] This step constructs an angle response feature matrix, comprehensively considering the influence of multiple physical factors such as basic reflection intensity, incident angle cosine effect, and optical path distance compensation, so that the extracted features have stronger physical significance. Furthermore, by calculating the three statistical features of average reflection intensity, reflection intensity standard deviation, and dynamic range, the reflection characteristics of the sample are fully characterized.

[0096] S2: Perform non-subsampled contourlet decomposition on each image in the reflected light intensity feature image sequence to obtain non-subsampled contourlet coefficients, calculate the local energy characteristics and directional consistency of the non-subsampled contourlet coefficients; calculate the adaptive fusion weights based on the local energy characteristics and directional consistency; fuse the adaptive fusion weights with the non-subsampled contourlet coefficients to obtain the fused non-subsampled contourlet coefficients; perform inverse transformation on the fused non-subsampled contourlet coefficients to obtain the comprehensive reflection feature map:

[0097] S21: Reflected light intensity feature image sequence Each image in Perform a non-subsampled Contourlet transform decomposition:

[0098] ;

[0099] in, For the The feature image is Tier Pixels in each direction The non-subsampled contourlet coefficients at ; is the non-subsampled contourlet transform decomposition function; for In pixels The pixel value at ; is the number of decomposition layers; For the The number of directions of the layer, ;

[0100] S22: Calculate the local energy characteristics and directional consistency of each non-subsampled contourlet coefficient:

[0101] ;

[0102] in, and They are The local energy characteristics and directional consistency of is the Gaussian weight function, , is a natural constant, and are the horizontal and vertical indices of the Gaussian kernel weight function, is the local window radius; For the The feature image is Tier Pixels in each direction The non-subsampled contourlet coefficients at ;

[0103] S23: Based on the local energy characteristics and directional consistency, the adaptive fusion weight is calculated; the adaptive fusion weight is fused with the non-subsampled contourlet coefficient to obtain the fused non-subsampled contourlet coefficient; the fused non-subsampled contourlet coefficient is inversely transformed to obtain the comprehensive reflection feature map:

[0104] S231: Calculate adaptive fusion weights based on local energy features and directional consistency:

[0105] ;

[0106] in, for Adaptive fusion weights of and They are The local energy characteristics and directional consistency of For the The feature image is Tier Pixels in each direction The non-subsampled contourlet coefficients at ;

[0107] S232: Fusing the adaptive fusion weights with the non-subsampled contourlet coefficients to obtain fused non-subsampled contourlet coefficients:

[0108] ;

[0109] in, After fusion Tier Pixels in each direction The non-subsampled contourlet coefficients at ;

[0110] S233: Perform an inverse transformation on the fused non-subsampled contourlet coefficients to obtain a comprehensive reflection feature map, specifically:

[0111] ;

[0112] in, Comprehensive reflection feature map In pixels The pixel value at ; is the non-subsampled contourlet inverse transform function.

[0113] This step uses non-subsampled contourlet transform technology to perform multi-scale decomposition and fusion of reflected light intensity features. Its technical advantages are mainly reflected in the three levels of feature decomposition, feature representation and feature fusion. In terms of feature decomposition, non-subsampled contourlet transform overcomes the limitations of traditional wavelet transform in directional selectivity and translation invariance, and achieves a fine characterization of image reflection features through a multi-level and multi-directional decomposition strategy. In particular, the use of a non-subsampled mechanism effectively avoids the frequency aliasing phenomenon in the traditional subsampling process and maintains the integrity and continuity of edge features.

[0114] In terms of feature characterization, this step comprehensively describes the spatial distribution characteristics of the reflection features through two key indicators: local energy features and directional consistency. Among them, the local energy features reflect the concentration of reflection intensity at a specific scale and direction, while the directional consistency quantifies the structural similarity of the local area. By introducing the Gaussian weight function for local feature calculation, the stability of feature extraction is guaranteed and the ability to suppress noise is enhanced. This dual feature characterization mechanism enables this method to more accurately capture the subtle reflection differences on the surface of the tempered glass film.

[0115] At the feature fusion level, this step designs an adaptive weight fusion strategy based on local energy features and directional consistency. This strategy can automatically adjust the fusion weight according to the importance of features at different scales and directions, ensuring that the final comprehensive reflection feature map retains important detail information while suppressing redundancy and noise components. The comprehensive feature map reconstructed by inverse transformation not only maintains the physical meaning of the original reflection features, but also improves the feature discrimination ability.

[0116] S3: Build a deep learning feature extraction network, extract features from the reflected light intensity feature image sequence and the comprehensive reflection feature map, and perform feature fusion to obtain the fused feature map:

[0117] S31: Construct two convolutional neural networks with the same structure (the first convolutional neural network and the second convolutional neural network) to respectively analyze the reflected light intensity feature image sequence and comprehensive reflection characteristic map Perform feature extraction and generate the first convolution feature map respectively And the second convolutional feature map :

[0118] ;

[0119] in, and They are the first convolution feature map and the second convolution feature map respectively; and They are the first convolutional neural network and the second convolutional neural network respectively. Each network contains 5 convolutional layers, and the convolution kernel size of each layer is ;

[0120] S32: Perform feature fusion:

[0121] ;

[0122] in, is the fused feature map In pixels The eigenvalue at ; and They are the first convolutional feature maps respectively. And the second convolutional feature map In pixels The eigenvalue at ; is the adaptive weight coefficient at the pixel point The coefficient value at is:

[0123] ;

[0124] in, and is a trainable weight matrix.

[0125] This step uses two independent but symmetrical five-layer convolutional neural networks. Each network extracts features through a small 3×3 convolution kernel. This design can effectively capture local texture information and reduce computational complexity. The five-layer network depth design enables the model to have sufficient nonlinear expression capabilities and can extract multi-level information from low-level texture features to high-level semantic features layer by layer.

[0126] S4: Use the feature pyramid network to extract the multi-scale features of the fused feature map, and perform feature aggregation to obtain the aggregated multi-scale features:

[0127] S41: Construct feature pyramid:

[0128] ;

[0129] in, For the pyramid levels; For the Layer feature pyramid at pixel point The eigenvalue at ; and They are upsampling and downsampling operations respectively; For the Convolution operation of the layer;

[0130] S42: Perform feature aggregation:

[0131] ;

[0132] in, The multi-scale features after aggregation are at the pixel point The eigenvalue at ; For the The weight coefficient of the layer feature.

[0133] By constructing a four-layer pyramid structure, this step realizes the multi-scale analysis of the reflection characteristics of the tempered glass film. Each layer of the pyramid gradually expands the receptive field through downsampling and convolution operations, allowing the network to simultaneously focus on the reflection feature information of different scales. This hierarchical feature extraction method is particularly suitable for processing the multi-scale reflection characteristics existing on the surface of the tempered glass film, and can fully express everything from the microscopic local texture to the macroscopic overall reflection pattern.

[0134] S5: Map the aggregated multi-scale features based on the fully connected network, and obtain the anti-reflection performance evaluation score through the activation function:

[0135] S51: Constructing the fully connected layer:

[0136] ;

[0137] in, is the feature vector; It is a fully connected layer operation; For the flattening operation;

[0138] S52: Calculate the evaluation score:

[0139] ;

[0140] in, is the anti-reflection performance evaluation score, ranging from ; is the activation function; and The weight matrix and bias term are evaluated for anti-reflection performance, respectively.

[0141] Embodiment 2: The present invention also discloses a mobile phone tempered glass film anti-reflection performance evaluation system, comprising the following five modules:

[0142] High dynamic range imaging module: uses a multi-angle light source array to illuminate the surface of the tempered glass film, obtains image sequences under different light sources through high dynamic range imaging technology, and calculates the reflected light intensity characteristic image sequence;

[0143] Non-subsampled contourlet decomposition module: perform non-subsampled contourlet decomposition on each image in the reflected light intensity feature image sequence to obtain non-subsampled contourlet coefficients, calculate the local energy characteristics and directional consistency of the non-subsampled contourlet coefficients; calculate the adaptive fusion weights based on the local energy characteristics and directional consistency; fuse the adaptive fusion weights with the non-subsampled contourlet coefficients to obtain the fused non-subsampled contourlet coefficients; perform inverse transformation on the fused non-subsampled contourlet coefficients to obtain a comprehensive reflection feature map;

[0144] Feature extraction module: build a deep learning feature extraction network, extract features from the reflected light intensity feature image sequence and the comprehensive reflection feature map, and perform feature fusion to obtain the fused feature map;

[0145] Feature aggregation module: The feature pyramid network is used to extract the multi-scale features of the fused feature map, and then feature aggregation is performed to obtain the aggregated multi-scale features;

[0146] Anti-reflection performance evaluation module: Maps the aggregated multi-scale features based on a fully connected network, and obtains the anti-reflection performance evaluation score through the activation function.

[0147] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0148] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0149] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for evaluating the anti-reflection performance of a mobile phone tempered glass film, characterized in that: The following steps are involved: S1: Use a multi-angle light source array to illuminate the surface of the tempered glass film, obtain image sequences under different light sources through high dynamic range imaging technology, and calculate the reflected light intensity characteristic image sequence; S2: Perform non-subsampled contourlet decomposition on each image in the reflected light intensity feature image sequence to obtain non-subsampled contourlet coefficients, and calculate the local energy characteristics and directional consistency of the non-subsampled contourlet coefficients; Based on the local energy characteristics and directional consistency, the adaptive fusion weight is calculated; the adaptive fusion weight is fused with the non-subsampled contourlet coefficient to obtain the fused non-subsampled contourlet coefficient; the fused non-subsampled contourlet coefficient is inversely transformed to obtain the comprehensive reflection feature map; S3: Build a deep learning feature extraction network, extract features from the reflected light intensity feature image sequence and the comprehensive reflection feature map, and perform feature fusion to obtain the fused feature map; S4: Use the feature pyramid network to extract the multi-scale features of the fused feature map, and perform feature aggregation to obtain the aggregated multi-scale features; S5: Map the aggregated multi-scale features based on the fully connected network, and obtain the anti-reflection performance evaluation score through the activation function.

2. The method for evaluating the anti-reflection performance of a mobile phone tempered glass film according to claim 1, characterized in that: The step S1 includes the following steps: S11: Construct a circular array of 8 LED light sources, with the light sources evenly distributed along the circumference. is the light source number, and the The azimuth of the light source and The incident angle of the light source ; S12: Acquire image sequences under different light sources ,in For the Images acquired under illumination of light sources, using high dynamic range imaging technology, with three exposure times collected at each light source position The following image is synthesized into a high dynamic range image: ； in, For the The image obtained under the illumination of a light source is at the pixel point The pixel value at ; are the horizontal and vertical coordinates of the image; For the Under the light source Exposure time The collected image is at pixel point The pixel value at ; is the weight function, specifically: ； in, and Respectively Under the light source Exposure time The minimum and maximum pixel values ​​of all pixels of the collected image; S13: Constructing the angular response feature matrix And calculate the reflected light intensity characteristic image sequence .

3. The method for evaluating the anti-reflection performance of a mobile phone tempered glass film according to claim 2, characterized in that: The step S13 includes the following steps: S131: Calculate the The angle response characteristic matrix at the pixel point under the illumination of the light source The response value at : ； in, Pixel To The distance between the projection centers of the light sources; is the focal length of the imaging system; S132: Calculate the statistical characteristics of the reflected light intensity of each pixel: ； in, is the average reflection intensity feature In pixels The pixel value at ; is the reflection intensity standard deviation characteristic In pixels The pixel value at ; is the reflection intensity dynamic range characteristic In pixels The pixel value at ; and They are , and The maximum and minimum values ​​in the image sequence of reflected light intensity characteristics .

4. The method for evaluating the anti-reflection performance of a mobile phone tempered glass film according to claim 2, characterized in that: The step S2 includes the following steps: S21: Reflected light intensity feature image sequence Each image in Perform a non-subsampled Contourlet transform decomposition: ； in, For the The feature image is Tier Pixels in each direction The non-subsampled contourlet coefficients at ; is the non-subsampled contourlet transform decomposition function; for In pixels The pixel value at ; is the number of decomposition layers; For the The number of directions of the layer, ; S22: Calculate the local energy characteristics and directional consistency of each non-subsampled contourlet coefficient: ； in, and They are The local energy characteristics and directional consistency of is the Gaussian weight function, , is a natural constant, and are the horizontal and vertical indices of the Gaussian kernel weight function, is the local window radius; For the The feature image is Tier Pixels in each direction The non-subsampled contourlet coefficients at ; S23: Based on the local energy characteristics and directional consistency, the adaptive fusion weights are calculated; the adaptive fusion weights are fused with the non-subsampled contour wave coefficients to obtain the fused non-subsampled contour wave coefficients; the fused non-subsampled contour wave coefficients are inversely transformed to obtain a comprehensive reflection feature map.

5. The method for evaluating the anti-reflection performance of a mobile phone tempered glass film according to claim 4, characterized in that: The step S23 includes the following steps: S231: Calculate adaptive fusion weights based on local energy features and directional consistency: ； in, for Adaptive fusion weights of and They are The local energy characteristics and directional consistency of For the The feature image is Tier Pixels in each direction The non-subsampled contourlet coefficients at ; S232: Fusing the adaptive fusion weights with the non-subsampled contourlet coefficients to obtain fused non-subsampled contourlet coefficients: ； in, After fusion Tier Pixels in each direction The non-subsampled contourlet coefficients at ; S233: Perform an inverse transformation on the fused non-subsampled contourlet coefficients to obtain a comprehensive reflection feature map, specifically: ； in, The comprehensive reflection feature map In pixels The pixel value at ; is the non-subsampled contourlet inverse transform function.

6. The method for evaluating the anti-reflection performance of a mobile phone tempered glass film according to claim 4, characterized in that: The step S3 includes the following steps: S31: Construct two convolutional neural networks with the same structure to analyze the reflected light intensity feature image sequence respectively. and comprehensive reflection characteristic map Perform feature extraction and generate the first convolution feature map respectively And the second convolutional feature map ; Each convolutional neural network consists of 5 convolutional layers, using Convolution kernel is used for feature extraction; S32: Perform feature fusion: ； in, is the fused feature map In pixels The eigenvalue at ; and They are the first convolutional feature maps respectively. And the second convolutional feature map In pixels The eigenvalue at ; is the adaptive weight coefficient at the pixel point The coefficient value at is: ； in, and is a trainable weight matrix.

7. The method for evaluating the anti-reflection performance of a mobile phone tempered glass film according to claim 6, characterized in that: The step S4 includes the following steps: S41: Construct feature pyramid: ； in, For the pyramid levels; For the Layer feature pyramid at pixel point The eigenvalue at ; and They are upsampling and downsampling operations respectively; For the Convolution operation of the layer; S42: Perform feature aggregation: ； in, The multi-scale features after aggregation are at the pixel point The eigenvalue at ; For the The weight coefficient of the layer feature.

8. The method for evaluating the anti-reflection performance of a mobile phone tempered glass film according to claim 7, characterized in that: The step S5 includes the following steps: S51: Constructing the fully connected layer: ； in, is the feature vector; It is a fully connected layer operation; For the flattening operation; S52: Calculate the evaluation score: ； in, is the anti-reflection performance evaluation score, ranging from ; is the activation function; and The weight matrix and bias term are evaluated for anti-reflection performance, respectively.

9. A mobile phone tempered glass film anti-reflection performance evaluation system, characterized in that: include: High dynamic range imaging module: uses a multi-angle light source array to illuminate the surface of the tempered glass film, obtains image sequences under different light sources through high dynamic range imaging technology, and calculates the reflected light intensity characteristic image sequence; Non-subsampled contourlet decomposition module: Perform non-subsampled contourlet decomposition on each image in the reflected light intensity feature image sequence to obtain non-subsampled contourlet coefficients, and calculate the local energy characteristics and directional consistency of the non-subsampled contourlet coefficients; Based on the local energy characteristics and directional consistency, the adaptive fusion weight is calculated; the adaptive fusion weight is fused with the non-subsampled contourlet coefficient to obtain the fused non-subsampled contourlet coefficient; the fused non-subsampled contourlet coefficient is inversely transformed to obtain the comprehensive reflection feature map; Feature extraction module: build a deep learning feature extraction network, extract features from the reflected light intensity feature image sequence and the comprehensive reflection feature map, and perform feature fusion to obtain the fused feature map; Feature aggregation module: The feature pyramid network is used to extract the multi-scale features of the fused feature map, and then feature aggregation is performed to obtain the aggregated multi-scale features; Anti-reflection performance evaluation module: maps the aggregated multi-scale features based on a fully connected network, and obtains the anti-reflection performance evaluation score through the activation function; To realize the anti-reflection performance evaluation method of a mobile phone tempered glass film as described in any one of claims 1-8.

Images