Organic electroluminescent device fault detection system and method based on image processing

Abstract

The invention provides an organic electroluminescent device fault detection system and method based on image processing, and relates to the technical field of organic electroluminescent device fault detection. The system comprises an image acquisition module, an image preprocessing module, a feature extraction module and a fault detection module. The image acquisition module acquires an excited-state light-emitting image of the organic light-emitting device; the image preprocessing module preprocesses the excited-state light-emitting image to obtain a preprocessed image; a feature extraction module performs feature extraction on the preprocessed image to obtain a fusion feature vector; the fault detection module constructs a channel-space two-dimensional attention mechanism and identifies fault classification and positioning information of the organic light-emitting device. According to the method, through efficient image preprocessing, intelligent feature extraction and accurate classification decision, the automation and intelligence level of fault detection is improved, the labor cost is reduced, the false detection and omission ratio is reduced, and powerful technical support and guarantee are provided for production of organic electroluminescent devices.

Description

Technical Field

[0001] The present invention relates to the technical field of organic electroluminescent device fault detection, and particularly to an organic electroluminescent device fault detection system based on image processing and an organic electroluminescent device fault detection method based on image processing. Background Art

[0002] Organic electroluminescent devices are a kind of optoelectronic display technology based on organic materials, and are widely used in multiple fields such as displays, televisions, and lighting. Compared with traditional liquid crystal display technology, organic electroluminescent devices have the advantages of self-luminescence, high contrast, wide viewing angle, and ultra-thin design, and thus are increasingly favored in modern electronic products. The basic working principle of organic electroluminescent devices is to utilize the electroluminescence effect generated when current passes through organic materials. When current passes through a structure sandwiching an organic layer, charge carriers will recombine in the light-emitting layer, releasing energy and emitting light in the form of photons. According to the different organic materials used, organic electroluminescent devices can emit light of different colors, and usually achieve full-color display by combining organic electroluminescent devices of different colors. The development of organic electroluminescent device technology began in the 1980s, and the initial research focused on small-sized displays such as mobile phones and digital photo frames. With the progress of material science and manufacturing processes, the manufacture of large-sized organic electroluminescent device displays has become possible, enabling organic electroluminescent devices to be used in televisions and large advertising display screens. Currently, organic electroluminescent devices have rapidly become the mainstream of television and smartphone display technologies, mainly due to their excellent picture quality and design flexibility.

[0003] After long-term use, organic electroluminescent devices may exhibit light attenuation, especially in the blue light part, which makes the light output uneven and affects the display effect. During the production process of organic electroluminescent devices, some tiny faults often occur, such as bubbles, foreign objects, or discontinuous light-emitting layers. These faults may not be apparent in the initial tests, but will cause performance degradation during subsequent use. Therefore, it is particularly important to detect and locate these faults in a timely manner. Some existing detection technologies cannot meet the requirements of modern production for efficient and high-precision detection. Summary of the Invention

[0004] The present invention provides an organic electroluminescent device fault detection system and method based on image processing to solve the defects existing in the prior art.

[0005] On the one hand, the present invention provides an organic electroluminescent device fault detection system based on image processing, including: An image acquisition module for acquiring the excited-state luminescence image of the organic electroluminescent device by using multi-spectral imaging technology.

[0006] An image preprocessing module is used to establish a static background model of the organic light-emitting device, eliminate the environmental light interference of the excited-state luminescence image, and perform local contrast enhancement on the excited-state luminescence image through the CLAHE algorithm to obtain a preprocessed image.

[0007] A feature extraction module is used to perform fused feature extraction and discrete wavelet transform on the preprocessed image, and perform principal component analysis to output a fused feature vector.

[0008] A fault detection module is used to construct a channel-spatial two-dimensional attention mechanism and identify the fault classification and location information of the organic light-emitting device based on the fused feature vector.

[0009] According to an organic light-emitting device fault detection system based on image processing provided by the present invention, the image acquisition module includes an illumination unit, an optical imaging unit, a multispectral sensor unit, and a data acquisition unit. The illumination unit is used to provide an adjustable light source to excite the organic light-emitting device. The optical imaging unit is used to collect the luminescence image of the organic light-emitting device through a camera sensor. The multispectral sensor unit is used to convert the luminescence image into a digital signal supporting multispectral imaging technology. The data acquisition unit is used to perform preliminary processing on the digital signal to obtain an excited-state luminescence image, and the preliminary processing includes image formatting, calibration, and time synchronization.

[0010] According to an organic light-emitting device fault detection system based on image processing provided by the present invention, the process of converting the luminescence image into a digital signal supporting multispectral imaging technology includes: Select a sensor with spectral response characteristics to convert the luminescence image into an electrical signal.

[0011] An analog-to-digital converter is used to sample and quantize the electrical signal to form a discrete digital signal.

[0012] The discrete digital signals are combined to form a multispectral data set to obtain a digital signal.

[0013] According to an organic light-emitting device fault detection system based on image processing provided by the present invention, the image preprocessing module includes a dynamic background modeling unit, a CLAHE enhancement unit, and a noise filtering unit. The dynamic background modeling unit is used to collect a standard background image of the organic light-emitting device to establish a static background model, and compare the pixel differences between the excited-state luminescence image and the static background model in real time, and determine the binary mask image of the candidate fault area by setting a dynamic threshold. The CLAHE enhancement unit is used to perform local contrast enhancement processing on the binary mask image to obtain a preprocessed image. The noise filtering unit is used to suppress noise on the preprocessed image by using a non-local means filtering algorithm.

[0014] According to an organic electroluminescent device fault detection system based on image processing provided by the present invention, the process of performing local contrast enhancement processing on a binary mask image includes: dividing the binary mask image into pixel patches, performing expansion processing on each pixel patch, where the expansion processing includes performing histogram equalization to expand the gray value distribution range and limiting the upper limit of contrast stretching, and splicing the pixel patches after the expansion processing into a complete image.

[0015] According to an organic electroluminescent device fault detection system based on image processing provided by the present invention, the feature extraction module includes a multi-scale convolution unit, a wavelet transform unit, and a feature encoding unit. The multi-scale convolution unit is used to construct a multi-scale convolutional neural network to extract a multi-scale fusion feature map of the preprocessed image. The wavelet transform unit is used to perform discrete wavelet transform on the multi-scale fusion feature map to obtain low-frequency energy features and high-frequency detail features. The feature encoding unit is used to perform principal component analysis on the low-frequency energy features and high-frequency detail features to obtain a fusion feature vector.

[0016] According to an organic electroluminescent device fault detection system based on image processing provided by the present invention, the process of performing principal component analysis on the low-frequency energy features and high-frequency detail features includes: Performing data standardization processing on the low-frequency energy features and high-frequency detail features to obtain standardized low-frequency features and standardized high-frequency features.

[0017] Calculating the covariance of the standardized low-frequency features and standardized high-frequency features to obtain a covariance matrix, where each element in the covariance matrix represents the covariance between the standardized low-frequency features and standardized high-frequency features.

[0018] Performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues and corresponding eigenvectors.

[0019] Setting a contribution rate threshold, calculating the cumulative variance contribution rate of each eigenvalue, and retaining the eigenvalues whose cumulative variance contribution rate reaches the contribution rate threshold to obtain a fusion feature vector.

[0020] According to an organic electroluminescent device fault detection system based on image processing provided by the present invention, the fault detection module includes a channel attention unit, a spatial attention unit, and a classification decision unit. The channel attention unit is used to perform global pooling on the fusion feature vector, allocate channel importance weights, and obtain a channel-weighted feature vector. The spatial attention unit is used to perform element-wise superposition on the channel-weighted feature vector and the fusion feature vector to obtain a spatially enhanced feature vector. The classification decision unit is used to receive the spatially enhanced feature vector, parse the fault type through a fully connected classifier, and generate a fault coordinate map in combination with the fusion feature vector to obtain an output classification result and location information.

[0021] According to an organic electroluminescent device fault detection system based on image processing provided by the present invention, the fully connected classifier includes an input layer, a plurality of fully connected layers, and an output layer. The input layer is used to receive the spatially enhanced feature vector. Each fully connected layer calculates the weight connection of the spatially enhanced feature vector to the next layer and introduces non-linearity through an activation function to obtain a fault classification. The output layer is used to convert the fault classification into a probability distribution.

[0022] On the other hand, the present invention also provides an organic electroluminescent device fault detection method based on image processing, including: Using multi-spectral imaging technology to photograph the organic electroluminescent device to obtain its excited-state luminescence image.

[0023] Establish a static background model of the organic electroluminescent device, and compare the pixel differences between the excited-state luminescence image and the static background model to obtain a binary mask image.

[0024] Apply the CLAHE algorithm to enhance the local contrast of the binary mask image to obtain a preprocessed image.

[0025] Perform fusion feature extraction and discrete wavelet transform on the preprocessed image to obtain low-frequency energy features and high-frequency detail features.

[0026] Perform principal component analysis on the low-frequency energy features and high-frequency detail features to obtain a fusion feature vector.

[0027] Construct a channel-spatial two-dimensional attention mechanism to detect faults of the organic electroluminescent device based on the fusion feature vector.

[0028] The organic electroluminescent device fault detection system and method based on image processing provided by the present invention can accurately distinguish the effective signals and noises in the electroluminescent device in a complex environment by introducing dynamic background modeling and multi-level feature extraction techniques, ensuring the minimum impact of background interference on the detection results. By integrating channel and spatial attention mechanisms, the system intelligently enhances the expression ability of key features, enabling it to freely select the most representative information for analysis when processing multi-scale features and improving the saliency of fault features. By using methods such as discrete wavelet transform and principal component analysis, the system accurately extracts low-frequency energy and high-frequency detail features, optimizes the feature dimension, reduces redundant information, and ensures the effectiveness and reliability of the feature vector. This not only improves the efficiency of the subsequent classifier but also enhances the generalization ability of the model under various working conditions. During the classification decision-making process, the ability to parse spatially enhanced features using a fully connected network and output the fault type and coordinate map further enhances the comprehensive recognition ability of the system, enabling it to quickly reflect the specific location and category of the fault for convenient real-time monitoring and subsequent processing. By combining traditional fault detection methods with advanced deep learning techniques through efficient image preprocessing, intelligent feature extraction, and precise classification decision-making, the automation and intelligence level of fault detection are significantly improved. This can not only reduce labor costs and the rate of false detections and missed detections but also ensure the stability and consistency of product quality while improving production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0030] Figure 1 It is a schematic structural diagram of the organic electroluminescent device fault detection system based on image processing provided by the embodiment of the present invention; Figure 2 It is a flowchart of the principal component analysis of the low-frequency energy feature and the high-frequency detail feature provided by this embodiment; Figure 3 It is a schematic flowchart of the organic electroluminescent device fault method based on image processing provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the protection scope of the present invention.

[0032] The following will Figures 1 - 3 describe the organic electroluminescent device fault detection system and method based on image processing of the present invention.

[0033] Figure 1 is a schematic structural diagram of the organic electroluminescent device fault detection system based on image processing provided by an embodiment of the present invention.

[0034] As Figure 1 shown, for the organic electroluminescent device fault detection system and method based on image processing provided by an embodiment of the present invention, the execution subject may be the organic electroluminescent device fault detection system based on image processing. The system includes an image acquisition module, an image preprocessing module, a feature extraction module, and a fault detection module.

[0035] The image acquisition module is used to acquire the excited-state luminescence image of the organic electroluminescent device by using multi-spectral imaging technology.

[0036] The image acquisition module includes an illumination unit, an optical imaging unit, a multi-spectral sensor unit, and a data acquisition unit. The illumination unit is used to provide an adjustable light source to excite the organic electroluminescent device. The optical imaging unit is used to acquire the luminescence image of the organic electroluminescent device through a camera sensor. The multi-spectral sensor unit is used to convert the luminescence image into a digital signal supporting multi-spectral imaging technology. The data acquisition unit is used to perform preliminary processing on the digital signal to obtain the excited-state luminescence image. The preliminary processing includes image formatting, calibration, and time synchronization.

[0037] The process of converting the luminescence image into a digital signal supporting multi-spectral imaging technology includes: Select a sensor with spectral response characteristics to convert the luminescence image into an electrical signal.

[0038] Use an analog-to-digital converter to sample and quantize the electrical signal to form a discrete digital signal.

[0039] Combine the discrete digital signals to form a multi-spectral data set to obtain the digital signal.

[0040] Select a sensor that can cover the emission wavelength band of the organic electroluminescent device. Consider the spectral sensitivity of the sensor to ensure that it can effectively detect the required light wavelength band. After the photons in the emission image enter the sensor, the photodiodes in the sensor sense the photons and generate corresponding currents. This current is proportional to the intensity of the incident light, so that light sources of different colors and intensities generate different electrical signal outputs. The analog-to-digital converter samples the electrical signals at a preset sampling rate. The sampling rate satisfies the Nyquist sampling theorem, that is, at least twice the highest frequency signal, so as to avoid aliasing. The sampled analog signal is converted into discrete values through the quantization algorithm of the analog-to-digital converter. The encodable level and precision of each sample are determined according to the resolution of the analog-to-digital converter, realizing the mapping of the continuous analog signal into discrete digital values. After sampling and quantization, the analog signal is converted into a digital signal. These digital signals can represent the comprehensive information of various intensities and wavelengths, forming preliminary emission data.

[0041] Multiple spectral signals obtained from the spectral response sensor will be transmitted to the data processing unit for data combination. The signals of each wavelength correspond to different characteristic information. The digital signals at each sampling point are integrated according to their wavelengths to form a multi-dimensional digital array. Each dimension corresponds to the optical signals of different wavelengths, constituting a complete multi-spectral data set. Check the integrity and validity of the multi-spectral data set, remove noise and invalid data, and ensure that the data set accurately represents the emission characteristics.

[0042] Convert the combined multi-spectral data into a format suitable for storage and processing. The processed and formatted multi-spectral digital signals will be handed over to the data acquisition unit for further analysis and processing to generate the final excited-state emission image.

[0043] The image preprocessing module is used to establish a static background model of the organic electroluminescent device, eliminate the environmental light interference of the excited-state emission image, and enhance the local contrast of the excited-state emission image through the CLAHE algorithm to obtain a preprocessed image.

[0044] The image preprocessing module includes a dynamic background modeling unit, a CLAHE enhancement unit, and a noise filtering unit. The dynamic background modeling unit is used to collect the standard background image of the organic electroluminescent device to establish a static background model, and compare the pixel differences between the excited-state emission image and the static background model in real time, and determine the binary mask image of the candidate fault area by setting a dynamic threshold. The CLAHE enhancement unit is used to perform local contrast enhancement processing on the binary mask image to obtain a preprocessed image. The noise filtering unit is used to suppress the noise of the preprocessed image by using the non-local means filtering algorithm.

[0045] The process of performing local contrast enhancement on a binary mask image includes: dividing the binary mask image into pixel patches, performing an expansion process on each pixel patch, where the expansion process includes performing histogram equalization to expand the range of gray values and limiting the upper limit of contrast stretching, and stitching the pixel patches after the expansion process into a complete image.

[0046] Regularly collect standard background images of the organic electroluminescent device when it is not excited. These images are used to construct a background model. Using a series of standard background images, a comprehensive static background model is constructed through an algorithm to reduce noise and random fluctuations in the images and ensure that the background model is as stable as possible. After exciting the organic electroluminescent device and obtaining the current image, the dynamic background modeling unit will compare the excited-state luminescence image with the static background model. By comparing pixel by pixel, the difference value is calculated. Calculate the mean and standard deviation of the background model, set an appropriate threshold, and generate a preliminary fault detection mask. Based on the calculated pixel differences and dynamic thresholds above, a binary mask image is generated, where potential fault regions are marked as 1 (or white), and the remaining background part is 0 (or black).

[0047] Divide the binary mask image into small patches of 8x8 or 16x16 pixels for processing. Apply the histogram equalization algorithm to calculate the gray histogram for each small patch to expand the range of gray values. Make the brightness distribution within the small patch more uniform and improve the contrast of the image. Use an adaptive method to limit the range of gray values, set an upper limit value to control the enhancement of contrast, and avoid noise amplification caused by over-enhancement. Re-stitch the processed small patches into a complete image to form an enhanced preprocessed image. At this time, the fault features in the image are more obvious, which helps subsequent analysis and processing.

[0048] Non-local mean filtering determines the final value of each pixel in the image by calculating the similarity between each pixel and all surrounding pixels. Compared with traditional local filtering methods, non-local mean filtering considers the context information of the entire image and can effectively retain image details while suppressing noise. For each pixel in the image, according to a certain similarity metric, such as Euclidean distance or cosine similarity, identify the pixels in its neighborhood that are similar to it. By selecting an appropriate search window, calculate the weighted average of the similar pixels. Multiply the values of the similar pixels by the weights and sum them to obtain the new value of each pixel. The weights are calculated based on the similarity, and pixels with higher similarity have higher weights. After completing the non-local mean filtering calculation for all pixels, a preprocessed image with noise suppression is generated. This image has a lower noise level, a smoother surface, and still retains important detail and boundary information.

[0049] The feature extraction module is used to perform fused feature extraction and discrete wavelet transform on the preprocessed image, and perform principal component analysis to output a fused feature vector.

[0050] The feature extraction module includes a multi-scale convolution unit, a wavelet transform unit, and a feature encoding unit. The multi-scale convolution unit is used to construct a multi-scale convolutional neural network to extract the multi-scale fusion feature map of the preprocessed image. The wavelet transform unit is used to perform discrete wavelet transform on the multi-scale fusion feature map to obtain the low-frequency energy feature and the high-frequency detail feature. The feature encoding unit is used to perform principal component analysis on the low-frequency energy feature and the high-frequency detail feature to obtain the fusion feature vector.

[0051] Figure 2 It is a flowchart for performing principal component analysis on the low-frequency energy feature and the high-frequency detail feature provided in this embodiment.

[0052] As Figure 2 shown, the process of performing principal component analysis on the low-frequency energy feature and the high-frequency detail feature includes: Perform data normalization processing on the low-frequency energy feature and the high-frequency detail feature to obtain the normalized low-frequency feature and the normalized high-frequency feature.

[0053] Calculate the covariance of the normalized low-frequency feature and the normalized high-frequency feature to obtain the covariance matrix. Each element in the covariance matrix represents the covariance between the normalized low-frequency feature and the normalized high-frequency feature.

[0054] Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues and the corresponding eigenvectors.

[0055] Set the contribution rate threshold, calculate the cumulative variance contribution rate of each eigenvalue, and retain the eigenvalues whose cumulative variance contribution rate reaches the contribution rate threshold to obtain the fusion feature vector.

[0056] Design a multi-layer convolutional neural network containing multiple convolutional layers and pooling layers. Each convolutional layer uses a different kernel size to capture features of different scales. Apply convolutional operations to each layer of the preprocessed image to extract the local features of specific regions of the image. The convolutional kernel slides over the entire image and calculates the feature values at each position. Apply an activation function after each convolutional layer to introduce non-linearity and help the network learn more complex features. Connect a pooling layer after the convolutional layer to reduce the dimension of the feature map. Pooling operations, such as max pooling or average pooling, can retain significant features, reduce the computational amount, and provide a certain degree of translational invariance. Concatenate or average the feature maps at different convolutional scales to form the final multi-scale fusion feature map.

[0057] Select different wavelet bases, such as Haar wavelets, approximation wavelets, etc., and input the multi-scale fusion feature map into the discrete wavelet transform method. Wavelet transform decomposes the image into multiple levels, separating the signal into low-frequency and high-frequency parts. The low-frequency part represents the basic contour of the image, while the high-frequency part contains detailed information. Select the low-frequency energy features to represent the main structural information and the high-frequency detail features to represent the texture and edge information as needed.

[0058] Use the Z-score normalization method to normalize the extracted low-frequency energy features and high-frequency detail features, transforming the features into a form with a mean of 0 and a standard deviation of 1. Calculate the covariance matrix of the normalized low-frequency features and normalized high-frequency features. This matrix shows the linear relationship between the two features. The elements of the matrix represent the covariance between the features. Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues and corresponding eigenvectors. The magnitude of the eigenvalue represents the amount of variance accounted for by the corresponding eigenvector. Set a contribution rate threshold (e.g., 95%), calculate the cumulative variance contribution rate of each eigenvalue to determine the number of features to be retained. According to the set contribution rate threshold, select the features that contribute more to the data variance, retain the eigenvectors of the eigenvalues, and ensure that the obtained fusion feature vector can effectively represent the information of the input features. Combine the selected eigenvectors into a fusion feature vector to form the final feature representation. The fusion feature vector contains key information, which is helpful for subsequent fault classification and recognition tasks. Output the generated fusion feature vector to the subsequent processing module to provide a basis for the final fault diagnosis and material identification.

[0059] The fault detection module is used to construct a channel-spatial two-dimensional attention mechanism to identify the fault classification and location information of organic electroluminescent devices based on the fusion feature vector.

[0060] The fault detection module includes a channel attention unit, a spatial attention unit, and a classification decision unit. The channel attention unit is used to perform global pooling on the fusion feature vector, assign channel importance weights, and obtain a channel-weighted feature vector. The spatial attention unit is used to perform element-wise addition on the channel-weighted feature vector and the fusion feature vector to obtain a spatially enhanced feature vector. The classification decision unit is used to receive the spatially enhanced feature vector, analyze the fault type through a fully connected classifier, and generate a fault coordinate map combined with the fusion feature vector to obtain the output classification result and location information.

[0061] The fully connected classifier includes an input layer, multiple fully connected layers, and an output layer. The input layer is used to receive the spatially enhanced feature vector. Each fully connected layer calculates the weight connection from the spatially enhanced feature vector to the next layer and introduces non-linearity through an activation function to obtain the fault classification. The output layer is used to convert the fault classification into a probability distribution.

[0062] Element-wise addition is performed on the channel-weighted feature vector and the original fused feature vector. While enhancing the importance of channels, the complete information of the fused features is maintained. A spatial attention map is generated by performing a convolution operation on the added features to calculate the spatial attention weights. The spatial attention weights reflect the importance of specific positions in the final feature representation. Element-wise multiplication is carried out between the calculated spatial attention map and the added features to generate a spatially enhanced feature vector, which contains enhanced information for high-importance regions and will be used for subsequent classification decisions.

[0063] Design an input layer to receive the spatially enhanced feature vector as the input to the subsequent neural network. Add multiple fully connected layers after the input layer. Each fully connected layer performs the following processing: Weight connection calculation: Each layer transforms the input feature vector to the next layer through a weight matrix. This weight matrix is trainable and updated through the backpropagation algorithm.

[0064] Nonlinear activation function: Apply an activation function, such as ReLU or Sigmoid, after each fully connected layer to introduce nonlinear features and capture more complex feature relationships.

[0065] The last layer is the output layer, whose function is to transform the feature vector entering this layer into a probability distribution for fault classification. Use the Softmax activation function to convert the output into a probability vector, where each element corresponds to the probability of a category.

[0066] Generate a coordinate map corresponding to potential fault locations by overlaying the attention map on the image or through a coordinate-based derivation process. Combine the spatially enhanced feature vector and the spatial attention map to calculate the coordinate information of the fault. Combine the output fault classification probability and the fault coordinate map to form the final decision output, comprehensively showing which regions have faults, as well as the categories of these faults and their importance probabilities. Output the final classification results, the type of each fault, and the corresponding location in a structured data format, such as JSON or XML, for convenient subsequent processing, visualization, or recording. The output information includes the type and probability value of each fault, and the coordinates of each fault, which are visually displayed by overlaying markers or bounding boxes on the image. The classification decision unit also provides interfaces so that other modules, such as the user interface module or the monitoring module, can conveniently obtain, visualize, or further process the output results, including passing the information to the data storage system or displaying it in real time.

[0067] In summary, this embodiment provides an organic electroluminescent device fault detection system based on image processing. By introducing dynamic background modeling and multi-level feature extraction techniques, the system can accurately distinguish the effective signals from the noise in the electroluminescent device in a complex environment, ensuring the minimum impact of background interference on the detection results. By integrating the channel and spatial attention mechanisms, the system intelligently enhances the expression ability of key features, enabling it to freely select the most representative information for analysis when processing multi-scale features and enhancing the saliency of fault features.

[0068] By using methods such as discrete wavelet transform and principal component analysis, the system accurately extracts low-frequency energy and high-frequency detail features, optimizes the feature dimension, reduces redundant information, and ensures the effectiveness and reliability of the feature vector. This not only improves the efficiency of the subsequent classifier but also enhances the generalization ability of the model under various working conditions. During the classification decision-making process, the system uses a fully connected network to analyze the spatially enhanced features and output the ability to identify fault types and coordinate maps, further enhancing the comprehensive recognition ability of the system, which can quickly reflect the specific location and category of the fault, facilitating real-time monitoring and subsequent processing.

[0069] Overall, this solution combines traditional fault detection methods with advanced deep learning techniques through efficient image preprocessing, intelligent feature extraction, and accurate classification decision-making, significantly improving the automation and intelligence levels of fault detection. It can not only reduce labor costs and the rates of missed detections and false alarms but also ensure the stability and consistency of product quality while increasing production efficiency, providing strong technical support and guarantee for the production of organic electroluminescent devices. Ultimately, it achieves the combination of high quality, high efficiency, and low cost, contributing to promoting the intelligent manufacturing process and technological innovation in related industries and enabling enterprises to gain greater advantages in the market competition.

[0070] Based on the same general inventive concept, the present invention also protects an organic electroluminescent device fault detection method based on image processing. The following describes the organic electroluminescent device fault detection method based on image processing provided by the present invention. The organic electroluminescent device fault detection method described below can be mutually referred to corresponding to the organic electroluminescent device fault detection system described above.

[0071] Figure 3 is a schematic flowchart of the organic electroluminescent device fault method based on image processing provided by the embodiment of the present invention.

[0072] As Figure 3 shown, the organic electroluminescent device fault method based on image processing includes: Using multi-spectral imaging technology to photograph the organic electroluminescent device to obtain its excited-state luminescence image.

[0073] A static background model of an organic light-emitting device is established, and a binary mask image is obtained by comparing the pixel differences between the excited-state luminescence image and the static background model.

[0074] The CLAHE algorithm is applied to enhance the local contrast of the binary mask image to obtain a preprocessed image.

[0075] Fusion feature extraction and discrete wavelet transform are performed on the preprocessed image to obtain low-frequency energy features and high-frequency detail features.

[0076] Principal component analysis is performed on the low-frequency energy features and high-frequency detail features to obtain a fusion feature vector.

[0077] A channel-spatial two-dimensional attention mechanism is constructed to detect faults of the organic light-emitting device based on the fusion feature vector.

[0078] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0079] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. Based on this understanding, the above technical solutions, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disc, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An organic electroluminescent device fault detection system based on image processing, characterized in that, Including: An image acquisition module, which is used to acquire the excited-state luminescence image of the organic electroluminescent device by using multi-spectral imaging technology; An image preprocessing module, which is used to establish a static background model of the organic electroluminescent device, eliminate the environmental light interference of the excited-state luminescence image, and perform local contrast enhancement on the excited-state luminescence image through the CLAHE algorithm to obtain a preprocessed image; A feature extraction module, which is used to perform fusion feature extraction and discrete wavelet transform on the preprocessed image, and perform principal component analysis to output a fusion feature vector; A fault detection module, which is used to construct a channel-space two-dimensional attention mechanism and identify the fault classification and location information of the organic electroluminescent device based on the fusion feature vector.

2. The organic electroluminescent device fault detection system based on image processing according to claim 1, wherein The image acquisition module includes an illumination unit, an optical imaging unit, a multi-spectral sensor unit, and a data acquisition unit; the illumination unit is used to provide an adjustable light source to excite the organic electroluminescent device; the optical imaging unit is used to collect the luminescence image of the organic electroluminescent device through a camera sensor; the multi-spectral sensor unit is used to convert the luminescence image into a digital signal supporting multi-spectral imaging technology; the data acquisition unit is used to perform preliminary processing on the digital signal to obtain an excited-state luminescence image, and the preliminary processing includes image formatting, calibration, and time synchronization.

3. The organic electroluminescent device fault detection system based on image processing according to claim 2, wherein The process of converting the luminescence image into a digital signal supporting multi-spectral imaging technology includes: Selecting a sensor with spectral response characteristics to convert the luminescence image into an electrical signal; Sampling and quantifying the electrical signal by using an analog-to-digital converter to form a discrete digital signal; Combining the discrete digital signals to form a multi-spectral data set to obtain a digital signal.

4. The organic electroluminescent device fault detection system based on image processing according to claim 1, wherein The image preprocessing module includes a dynamic background modeling unit, a CLAHE enhancement unit, and a noise filtering unit; The dynamic background modeling unit is used to collect a standard background image of the organic electroluminescent device to establish a static background model, and compare the pixel differences between the excited-state luminescence image and the static background model in real time, and determine the binary mask image of the candidate fault area by setting a dynamic threshold; the CLAHE enhancement unit is used to perform local contrast enhancement processing on the binary mask image to obtain a preprocessed image; The noise filtering unit is used to suppress noise on the preprocessed image by using a non-local means filtering algorithm.

5. The organic electroluminescent device fault detection system based on image processing according to claim 4, wherein, The process of performing local contrast enhancement processing on the binary mask image includes: dividing the binary mask image into pixel blocks, performing expansion processing on each pixel block, and the expansion processing includes performing histogram equalization to expand the gray value distribution range and limiting the upper limit of the contrast stretching, and splicing the expanded pixel blocks into a complete image.

6. The organic electroluminescent device fault detection system based on image processing according to claim 1, characterized in that, The feature extraction module includes a multi-scale convolution unit, a wavelet transform unit, and a feature encoding unit. The multi-scale convolution unit is used to construct a multi-scale convolutional neural network to extract the multi-scale fusion feature map of the preprocessed image. The wavelet transform unit is used to perform discrete wavelet transform on the multi-scale fusion feature map to obtain low-frequency energy features and high-frequency detail features. The feature encoding unit is used to perform principal component analysis on the low-frequency energy features and the high-frequency detail features to obtain a fusion feature vector.

7. The organic electroluminescent device fault detection system based on image processing according to claim 6, characterized in that, The process of performing principal component analysis on the low-frequency energy features and the high-frequency detail features includes: Performing data standardization processing on the low-frequency energy features and the high-frequency detail features to obtain standardized low-frequency features and standardized high-frequency features; Calculating the covariance of the standardized low-frequency features and the standardized high-frequency features to obtain a covariance matrix, where each element in the covariance matrix represents the covariance between the standardized low-frequency features and the standardized high-frequency features; Performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues and corresponding eigenvectors; Setting a contribution rate threshold, calculating the cumulative variance contribution rate of each eigenvalue, and retaining the eigenvalues whose cumulative variance contribution rate reaches the contribution rate threshold to obtain a fusion feature vector.

8. The organic electroluminescent device fault detection system based on image processing according to claim 1, wherein The fault detection module includes a channel attention unit, a spatial attention unit, and a classification decision unit. The channel attention unit is used to perform global pooling on the fusion feature vector, assign channel importance weights, and obtain a channel-weighted feature vector. The spatial attention unit is used to perform element-wise addition on the channel-weighted feature vector and the fusion feature vector to obtain a spatially enhanced feature vector; The classification decision unit is used to receive the spatially enhanced feature vector, analyze the fault type through a fully connected classifier, and generate a fault coordinate map based on the fusion feature vector to obtain an output classification result and location information.

9. The organic electroluminescent device fault detection system based on image processing according to claim 1, characterized in that The fully connected classifier includes an input layer, multiple fully connected layers, and an output layer. The input layer is used to receive the spatially enhanced feature vector. Each fully connected layer calculates the weight connection from the spatially enhanced feature vector to the next layer and introduces non-linearity through an activation function to obtain a fault classification. The output layer is used to convert the fault classification into a probability distribution.

10. A method for detecting faults in an organic light-emitting device based on image processing, which uses the system for detecting faults in an organic light-emitting device based on image processing according to any one of claims 1 to 9, characterized in that, The method for detecting faults in organic electroluminescent devices based on image processing includes: Using multi-spectral imaging technology to photograph an organic electroluminescent device to obtain its excited-state luminescence image; Establishing a static background model of the organic electroluminescent device, and comparing the pixel differences between the excited-state luminescence image and the static background model to obtain a binary mask image; Applying the CLAHE algorithm to enhance the local contrast of the binary mask image to obtain a preprocessed image; Performing fusion feature extraction and discrete wavelet transform on the preprocessed image to obtain low-frequency energy features and high-frequency detail features; Performing principal component analysis on the low-frequency energy features and the high-frequency detail features to obtain a fusion feature vector; Constructing a channel-spatial two-dimensional attention mechanism to detect faults in organic electroluminescent devices based on the fusion feature vector.

Images