Method and system for detecting defects of plastic packaging materials in radiation environment based on machine vision

Abstract

The present application relates to the field of image processing technology, and discloses a method and system for detecting defects in plastic packaging materials in a radiation environment based on machine vision. The method comprises: collecting radiation-resistant images in a gamma-ray environment through a high-radiation-tolerant image sensor shielded by lead-tungsten alloy, and performing adaptive contrast enhancement. Defect features, including microcracks, bubbles, deformation, and color difference, are extracted using dual-tree complex wavelet transform and HSV color space. Multi-dimensional features are input into the AR‑DCNN network for multi-task learning to identify and score defect types. Based on the Weibull distribution and random forest algorithm, the radiation resistance life of the plastic packaging material is predicted, and the material degradation curve and safety threshold warning are obtained. The present application realizes the standardized processing of quantitative indicators of radiation damage and effectively improves the prediction accuracy.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method and system for detecting defects in plastic packaging materials in a radiation environment based on machine vision. Background Art

[0002] In the modern aerospace and electronics fields, plastic encapsulation materials, as an important component of packaging materials, are widely used in the protection and packaging of electronic components, integrated circuits, and other equipment. Especially in high-radiation environments such as spacecraft and satellites, these materials are subject to long-term radiation damage. Existing technologies generally rely on traditional testing methods, such as manual inspection and physical testing, to evaluate the performance and radiation damage of plastic encapsulation materials. These methods often require a lot of time and labor costs, and it is difficult to obtain real-time and accurate damage information of materials in complex radiation environments. In addition, traditional methods have limited processing capabilities for large-scale data and are unable to effectively explore the potential relationship between damage and material properties, resulting in insufficient accuracy and efficiency.

[0003] Existing technical methods lack efficient data processing and intelligent evaluation methods in the analysis of quantitative indicators of radiation damage. Traditional test data processing methods mostly rely on manual experience and are unable to uniformly and standardize the processing of various different radiation damage indicators. In addition, existing methods usually rely on a single data source, resulting in incomplete data dimensions and difficulty in adapting to complex changes in the radiation environment. Therefore, existing technologies lack effective quantitative analysis tools for accurately evaluating the electrical properties, mechanical strength, and thermal stability of plastic packaging materials. Especially in large-scale applications, their limitations are becoming more and more significant, and they cannot meet the high requirements of modern spacecraft and electronic equipment for the performance prediction of plastic packaging materials. Summary of the Invention

[0004] The present application provides a method and system for detecting defects in plastic packaging materials in a radiation environment based on machine vision, which is used to achieve standardized processing of quantitative indicators of radiation damage and effectively improve prediction accuracy.

[0005] In the first aspect, the present application provides a method for detecting defects in plastic packaging materials in a radiation environment based on machine vision, and the method for detecting defects in plastic packaging materials in a radiation environment based on machine vision includes: performing image acquisition and adaptive contrast enhancement processing on the surface of the plastic packaging material of the spacecraft electronic component through a lead-tungsten alloy shielded high radiation tolerance image sensor in a gamma-ray environment to obtain a radiation-resistant enhanced image; performing dual-tree complex wavelet transform feature decomposition and HSV color space defect feature extraction on the radiation-resistant enhanced image to obtain multi-dimensional feature representation data including microcracks, bubbles, deformation, and color difference features; inputting the multi-dimensional feature representation data into the AR-DCNN network with residual connection and attention mechanism to perform void convolution multi-task learning to obtain defect type identification and severity scoring results; based on the defect type identification and severity scoring results, the radiation resistance life of plastic packaging materials for nuclear facilities is predicted by Weibull distribution fitting and random forest algorithm to obtain a material performance degradation curve and safety threshold warning.

[0006] In a second aspect, the present application provides a system for detecting defects in plastic packaging materials in a radiation environment based on machine vision, the system comprising:

[0007] The enhancement module is used to perform image acquisition and adaptive contrast enhancement processing on the surface of the plastic packaging material of the spacecraft electronic components under a gamma-ray environment through a lead-tungsten alloy shielded high radiation tolerance image sensor to obtain a radiation-resistant enhanced image;

[0008] An extraction module is used to perform dual-tree complex wavelet transform feature decomposition and HSV color space defect feature extraction on the anti-radiation enhanced image to obtain multi-dimensional feature representation data including microcracks, bubbles, deformation, and color difference features;

[0009] An input module is used to input the multi-dimensional feature representation data into the AR-DCNN network with residual connection and attention mechanism to perform void convolution multi-task learning to obtain defect type recognition and severity scoring results;

[0010] The prediction module is used to predict the radiation resistance life of plastic packaging materials for nuclear facilities based on the defect type identification and severity scoring results through Weibull distribution fitting and random forest algorithm, and obtain the material performance degradation curve and safety threshold warning.

[0011] In a third aspect, a device for detecting defects in plastic packaging materials in a radiation environment based on machine vision is provided, comprising: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the device for detecting defects in plastic packaging materials in a radiation environment based on machine vision executes the above-mentioned method for detecting defects in plastic packaging materials in a radiation environment based on machine vision.

[0012] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium. When the computer-readable storage medium is run on a computer, the computer executes the above-mentioned method for detecting defects in plastic packaging materials in a radiation environment based on machine vision.

[0013] In the technical solution provided by this application, in terms of the standardized processing of radiation damage quantitative indicators, this solution uses feature normalization processing to enable various indicators to be compared within a unified numerical range, providing a reliable data basis for subsequent intelligent prediction. The standardized damage index vector can eliminate the influence between different dimensions, so that the data has good comparability and consistency, and provides effective input data for the training of machine learning models. Secondly, a training set is constructed based on samples in the historical experimental database, and a comprehensive prediction of material properties is achieved by matching the standardized damage index vector with the samples. The random forest algorithm is used to perform integrated learning through multiple decision trees. Under diversified data input, overfitting can be effectively avoided, and the accuracy and robustness of the prediction can be improved. The advantage of the random forest algorithm is that it can process high-dimensional data and has a strong modeling ability for complex relationships between features. Therefore, it can more accurately predict multiple performance indicators such as electrical properties, mechanical strength and thermal stability of plastic packaging materials. In particular, for electrical performance prediction, an ensemble learning approach based on a set of decision trees can use the average prediction value of all decision trees as the final result, thereby improving the prediction accuracy of dielectric strength and volume resistivity. This is crucial for the long-term reliability of electronic components in spacecraft and high-radiation environments. Similarly, for mechanical performance and thermal stability prediction, the decision tree model effectively combines quantitative radiation damage indicators to provide accurate predictions of mechanical properties such as tensile strength, flexural strength, and hardness, as well as thermal stability indicators such as heat deformation temperature and thermal expansion coefficient. These indicators play a decisive role in the thermal reliability of spacecraft electronic components in radiation environments. Accurate predictions provide theoretical support and data basis for material optimization and application. This solution incorporates artificial intelligence algorithms, particularly the random forest model, which significantly improves the efficiency of data processing and model prediction. Compared to traditional manual testing and calculation methods, random forests, by processing multiple decision trees in parallel, can rapidly extract underlying patterns and patterns from complex data, avoiding the subjectivity and errors of manual evaluation. More importantly, as the number of samples increases, random forests gradually improve prediction accuracy, requiring minimal human intervention and demonstrating a high level of automation and intelligence. This feature is particularly important for spacecraft electronic components and related fields that require large-scale data processing. It can quickly respond to new experimental data and achieve real-time and accurate damage assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 Schematic diagram of an embodiment of a method for detecting defects in plastic packaging materials in a radiation environment based on machine vision in an embodiment of the present application;

[0016] Figure 2 Schematic diagram of an embodiment of a system for detecting defects in plastic packaging materials in a radiation environment based on machine vision in an embodiment of the present application;

[0017] Figure 3 The figure is a schematic block diagram of the structure of a device for detecting defects in plastic packaging materials in a radiation environment based on machine vision in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] An embodiment of the present application provides a method for detecting defects in plastic packaging materials in a radiation environment based on machine vision. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices.

[0019] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In one embodiment of the present application, a method for detecting defects in plastic packaging materials in a radiation environment based on machine vision includes:

[0020] Step S101: Using a lead-tungsten alloy shielded high-radiation-tolerant image sensor, the surface of the plastic packaging material of the spacecraft electronic component is subjected to image acquisition and adaptive contrast enhancement processing in a gamma-ray environment to obtain a radiation-resistant enhanced image.

[0021] Step S102: performing dual-tree complex wavelet transform feature decomposition and HSV color space defect feature extraction on the radiation-resistant enhanced image to obtain multi-dimensional feature representation data including microcracks, bubbles, deformation, and color difference features;

[0022] Step S103: Input the multi-dimensional feature representation data into the AR-DCNN network with residual connection and attention mechanism to perform dilated convolution multi-task learning to obtain defect type recognition and severity scoring results;

[0023] Step S104: Based on the defect type identification and severity score results, the radiation resistance life of the plastic packaging material for nuclear facilities is predicted by Weibull distribution fitting and random forest algorithm to obtain the material performance degradation curve and safety threshold warning.

[0024] It is understandable that the execution subject of this application can be a plastic packaging material defect detection system in a radiation environment based on machine vision, or a terminal or a server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0025] In an embodiment of the present application, images are captured and enhanced on the surface of the plastic encapsulation material of spacecraft electronic components. Plastic encapsulation materials of spacecraft electronic components are prone to microcracks and other defects in radiation environments, making these defects difficult to detect using conventional methods. This method utilizes a lead-tungsten alloy shielded image sensor with a high radiation tolerance. A lead-tungsten alloy composite shielding layer of a specified thickness is applied to the sensor's exterior, forming a radiation-resistant sensor with a high shielding ratio. The internal circuitry utilizes a radiation-resistant reinforcement scheme with multiple redundancy designs. Under gamma-ray radiation, an automatic parameter correction unit monitors the sensor's radiation exposure level in real time. When the radiation intensity exceeds a preset threshold, the camera sensitivity parameters and exposure time are automatically adjusted. The captured raw image is divided into multiple frequency subbands using wavelet decomposition. Adaptive threshold denoising based on the current radiation intensity is applied to the high-frequency subbands. The denoised image is then locally enhanced using an adaptive contrast enhancement algorithm with a fixed pixel window. The enhancement coefficient is dynamically determined based on the local entropy value of the image. Pattern matching is used to identify the type of interference in the initial enhanced image, and corresponding compensation strategies are applied to correct the interference, ultimately resulting in a radiation-resistant enhanced image.

[0026] Feature extraction is performed on radiation-resistant enhanced images. First, a geometric deformation correction algorithm is applied, and the image distortion caused by radiation is eliminated through a multi-control point bicubic B-spline transform model. Subsequently, a dual-tree complex wavelet transform is applied to the corrected image to decompose the image into multiple decomposition scales, each of which contains multiple directional subbands, resulting in multi-scale and multi-directional feature subbands. The dual-tree complex wavelet transform is a surface texture analysis method for plastic packaging materials unique to nuclear radiation environments and has a significant detection effect on radiation microcracks. Local second-order matrix features, Gabor texture features, and morphological features are extracted from these feature subbands to obtain structural feature vectors. Simultaneously, the corrected image is converted to the HSV color space, and the statistical features of the three channels of hue, saturation, and lightness are extracted to obtain a color feature vector. This process quantifies the characteristic discoloration of plastic packaging materials caused by gamma-ray radiation. The structural feature vector and the color feature vector are concatenated to form an initial feature vector. Dimensionality reduction and enhanced separability are achieved through principal component analysis and linear discriminant analysis. A nonlinear mapping function is then used to enhance the differences between defective and normal areas, constructing a multi-scale feature pyramid. This ultimately yields a multi-dimensional feature representation containing features of microcracks, bubbles, deformation, and color difference. This multi-dimensional feature representation is then fed into a specially designed deep learning network for processing. An encoder-decoder AR-DCNN (Anti-Radiation Deep Convolutional Neural Network) is constructed. The encoder consists of multiple encoding blocks, each with three parallel multi-scale convolutional units. A radiation feature enhancement module is implemented after the encoder. The multi-scale feature map is processed using a residual connection block and a channel-wise attention mechanism to generate an enhanced feature representation. The channel-wise attention mechanism is specifically designed to enhance the channel-wise response of the plastic encapsulation material features of spacecraft electronic components. The enhanced feature representation is processed through multiple dilated convolutional layers in the decoder to form a multi-scale receptive field. This feature map is then upsampled using transposed convolutions between the decoding blocks and fused with the feature maps from the corresponding encoding layers. The fused feature map is then fed into a multi-task learning head consisting of three parallel branches: classification, segmentation, and scoring. The classification head outputs a probability distribution of defect types through a fully connected layer. The segmentation head generates a segmentation mask for the defect region through a 1×1 convolution operation. The scoring head outputs a defect severity score through a fully connected layer after global average pooling. The model is optimized through a combined loss function, ultimately outputting defect type identification and severity scoring results.

[0027] A density-weighted Voronoi diagram was constructed to associate detected defects with surrounding areas, resulting in a radiation damage pattern map. A kernel density estimation algorithm was applied to calculate the damage distribution density function on the material surface, yielding the surface damage distribution characteristics of the plastic encapsulation material. Support vector regression was used to analyze the correlation between different types of defects and radiation exposure, distinguishing between defects caused by radiation and those caused by non-radiation factors, and obtaining quantitative indicators of radiation damage. These quantitative indicators were normalized and then input into a random forest algorithm to construct a performance degradation assessment model containing multiple decision trees. This model outputs predicted values ​​for key performance parameters such as electrical performance, mechanical strength, and thermal stability. These predicted values ​​were fitted with a Weibull distribution to establish a functional relationship between the performance degradation rate and time, yielding a material performance degradation curve. The Weibull distribution is particularly well-suited for describing failure modes of plastic encapsulation materials in gamma-ray radiation environments. Based on the performance degradation curve, the estimated time for performance to drop to a specified percentage of the initial value was determined, and multi-level safety thresholds were set to ultimately provide safety threshold warning information for the use of plastic encapsulation materials in nuclear facilities.

[0028] For example, the plastic encapsulation material of spacecraft communication components was exposed to a cumulative gamma-ray dose of 500 Gy. Images were captured using a high-radiation-tolerant image sensor. After adaptive contrast enhancement, the dual-tree complex wavelet transform was decomposed into six scales. HSV color space features were extracted, revealing a significant discoloration of the material due to a shift in the mean of the H channel by 0.15 from its initial value. This multidimensional feature representation was then fed into an AR-DCNN network, which identified two microcracks (with a type confidence of 0.92) and one bubble (with a type confidence of 0.87), each with a severity score of 7.5. Density-weighted Voronoi diagrams and kernel density estimation determined that the damage was primarily concentrated at the component edges. Support vector regression analysis revealed a correlation coefficient of 0.85 between cracks and radiation, indicating typical radiation-induced damage. A random forest algorithm predicted a drop in insulation strength to 81% of the initial value, while the heat deformation temperature decreased by 12°C. Weibull distribution fitting determined that the material performance degradation rate followed specific shape parameters and scale parameters. It was predicted that after another cumulative radiation dose of 300 Gy, the material performance would drop below the safety threshold. The system issued an early warning message, recommending that the plastic packaging material of the component be replaced before the next maintenance cycle.

[0029] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0030] A lead-tungsten alloy composite shielding layer of a specified thickness is placed on the exterior of the image sensor to create a radiation-resistant sensor with a high shielding ratio. A radiation-resistant reinforcement scheme with multiple redundant designs is implemented within the internal circuits of the radiation-resistant sensor to create a circuit system that maintains normal function in high-energy radiation environments.

[0031] The sensor's radiation exposure level is monitored in real time through the parameter automatic correction unit. When the radiation intensity exceeds the preset threshold, the camera's sensitivity parameters and exposure time are automatically adjusted to obtain image acquisition parameters suitable for the current radiation environment.

[0032] The surface image of the plastic packaging material is acquired according to the image acquisition parameters to obtain the image of the material under the original radiation environment;

[0033] The material image under the original radiation environment is decomposed into multiple frequency sub-bands by wavelet decomposition, and the adaptive threshold related to the current radiation intensity is applied to the high-frequency sub-band to obtain the denoised image;

[0034] An adaptive contrast enhancement algorithm with a fixed pixel window size and a set overlap ratio is applied to the denoised image for local enhancement. The enhancement coefficient is dynamically determined according to the local entropy value of the image to obtain a preliminary enhanced image.

[0035] According to the interference feature patterns of scattering, drift, and thermal noise radiation contained in the interference feature library, the interference type in the preliminary enhanced image is identified through pattern matching to obtain the image interference feature;

[0036] According to the image interference characteristics, the corresponding compensation strategy is applied to correct the interference of the preliminary enhanced image to obtain the final radiation-resistant enhanced image. At the same time, the current radiation type, radiation intensity, and cumulative dose are recorded to form radiation environment parameter data.

[0037] Specifically, applying a lead-tungsten alloy composite shielding layer to the image sensor is a key step in achieving a radiation-hardened sensor. This composite shielding layer is composed of two materials: lead, which provides excellent gamma-ray shielding capabilities, and tungsten, which offers excellent neutron shielding. In nuclear facilities, the thickness of the shielding layer is determined by the radiation intensity. In gamma-ray environments, the high-shielding efficiency of the lead-tungsten alloy composite shielding layer effectively isolates over 90% of radiation sources, protecting the sensor chip from radiation damage. A radiation-hardened sensor with multiple redundancies within the sensor's internal circuitry is crucial for maintaining proper functionality in high-energy radiation environments. This multiple-redundancy design utilizes a three-module redundant architecture, whereby the same function is executed in parallel by three independent circuit modules. A voting mechanism selects the majority result as the output, eliminating errors caused by single-event upsets (SEEs). The circuit system also utilizes radiation-hardened electronic components, including transistors with a specially doped process and integrated circuits with thickened gate oxides. An automatic parameter calibration unit monitors the sensor's radiation exposure level in real time via a built-in radiation detection module, consisting of a PIN diode array. When radiation intensity exceeds a preset threshold, the module automatically adjusts the camera's sensitivity parameters and exposure time. During the adjustment process, the sensitivity parameter increases or decreases based on the logarithmic ratio of the radiation intensity, while the exposure time is adjusted based on the inverse square root of the radiation intensity. These two parameters combine to form image acquisition parameters that are suitable for the current radiation environment, ensuring that sufficiently clear images can be obtained under different radiation intensities.

[0038] Capturing the surface image of the plastic encapsulating material according to image acquisition parameters involves obtaining an image of the material under its original radiation environment. During acquisition, the camera maintains a fixed distance from the surface of the plastic encapsulating material, and the lighting angle is set to 45 degrees to avoid strong reflections. The captured original image is a 16-bit grayscale image with a resolution of 2048 × 1536 pixels. This pixel depth provides sufficient information for subsequent detection of subtle defects.

[0039] The image of materials under the original radiation environment usually contains a large amount of radiation noise, which requires noise reduction through wavelet decomposition. Wavelet decomposition decomposes the image into multiple frequency subbands, including a low-frequency subband and three high-frequency subbands (horizontally, vertically and diagonally). For each decomposition level, the low-frequency subband is decomposed again to form a multi-layer wavelet decomposition tree. Adaptive threshold denoising related to the current radiation intensity is applied to the high-frequency subband. The threshold calculation formula is: threshold value = basic threshold + radiation intensity coefficient × current radiation intensity. Soft threshold processing is performed on the wavelet coefficients exceeding the threshold, that is, the coefficients exceeding the threshold are reduced by the threshold value to retain the continuity of the signal. After the inverse wavelet transform, the denoised image is obtained, which effectively removes the high-frequency noise introduced by the radiation environment.

[0040] Applying an adaptive contrast enhancement algorithm to the denoised image is a key step in enhancing the characteristics of defects in plastic packaging materials. This algorithm processes the image block by block using a sliding window with a fixed pixel window size (typically 16×16 pixels) and a 50% overlap ratio. The enhancement coefficient is dynamically determined based on the image's local entropy, which considers the probability distribution of pixel grayscale values ​​within the window. A larger enhancement coefficient is used for regions with high entropy (indicating high information content and possible defects), while a smaller enhancement coefficient is used for regions with low entropy (indicating low information content and possible background). This processing results in a preliminary enhanced image in which defect features are effectively highlighted.

[0041] The interference feature library is a unique image processing resource for radiation environments. It contains typical radiation interference signature patterns such as scattering, drift, and thermal noise. Scattering appears as star-shaped noise in the image, drift as a gradual change in overall image brightness, and thermal noise as randomly distributed bright spots. Template matching is used to calculate the similarity between the initial enhanced image and each pattern in the interference feature library, identifying the primary interference type in the current image and forming an image interference feature description. Applying appropriate compensation strategies based on the image interference characteristics is key to ultimately obtaining high-quality radiation-resistant enhanced images. For scattering interference, median filtering is applied to remove point noise; for drift interference, bilateral filtering is applied to preserve edges while smoothing brightness gradients; and for thermal noise, morphological opening is applied to remove random bright spots. After compensation, the final radiation-resistant enhanced image is obtained. The radiation type (gamma rays, X-rays, etc.), radiation intensity (Gy / h), and cumulative dose (Gy) are also recorded to form radiation environment parameter data. This data is transmitted along with the image to subsequent processing modules.

[0042] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0043] A geometric deformation correction algorithm is applied to the radiation-resistant enhanced image, and the image distortion caused by radiation is eliminated through a bicubic B-spline transformation model with multiple control points to obtain a corrected image with sub-pixel accuracy.

[0044] The dual-tree complex wavelet transform is applied to the rectified image to decompose the image into multiple decomposition scales. Each scale contains multiple directional sub-bands, resulting in multi-scale and multi-directional feature sub-bands. The dual-tree complex wavelet transform is a surface texture analysis method for plastic packaging materials unique to nuclear radiation environments and has a significant detection effect on radiation-induced microcracks.

[0045] The local second-order matrix features, Gabor texture features and morphological features are extracted from the multi-scale and multi-directional feature sub-bands to obtain the structural feature vectors of the spacecraft electronic component packaging materials.

[0046] The rectified image is converted to the HSV color space, and statistical features of the hue, saturation, and lightness channels are extracted, including mean, standard deviation, skewness, kurtosis, energy, and entropy, to obtain a color feature vector. The HSV color space feature extraction quantifies the characteristic discoloration of the plastic packaging material caused by gamma-ray radiation.

[0047] The structural feature vector and the color feature vector are concatenated to form the initial feature vector, and the feature dimension is reduced to a lower dimension through principal component analysis, retaining the main information to obtain the reduced dimension feature vector;

[0048] The reduced dimensionality feature vector is further reduced by linear discriminant analysis, while maximizing the separability between different defect categories to obtain the optimized feature vector;

[0049] The defect feature enhancement algorithm is applied to the optimized feature vector, and the difference between the defect area and the normal area in the feature space is enhanced through a nonlinear mapping function. A multi-scale feature pyramid is constructed to obtain multi-dimensional feature representation data including typical microcracks, bubbles, deformation, and color difference features in the nuclear facility environment.

[0050] Specifically, a geometric distortion correction algorithm is applied to eliminate radiation-induced image distortion. In high-radiation environments, images often experience radial and tangential distortion, manifesting as straight lines becoming curved or distorted. The geometric distortion correction algorithm uses a multi-control-point bicubic B-spline transform model to perform corrections. Specifically, the algorithm identifies feature points in the image, establishes a correspondence between these feature points and their ideal positions, and constructs a transformation matrix. The bicubic B-spline transform model uses 16 control points to influence the mapping relationship of each pixel, ensuring a smooth and continuous transformation. The correction process iteratively minimizes the Euclidean distance between the actual and ideal positions of the control points until sub-pixel accuracy is achieved (error less than 0.5 pixels). The resulting corrected image geometry is restored to a radiation-free state, laying the foundation for subsequent feature extraction. Applying the dual-tree complex wavelet transform to the corrected image is a key step in extracting multi-scale and multi-directional features. The dual-tree complex wavelet transform is a surface texture analysis method for plastic packaging materials unique to nuclear radiation environments. Compared with traditional wavelet transforms, it has increased directional selectivity, making it particularly suitable for detecting radiation-induced microcracks. This transform uses two sets of orthogonal wavelet filters, one for the real part and the other for the imaginary part, which together form a complex wavelet. In actual calculations, the image is decomposed at multiple levels, with each level generating low-frequency and high-frequency subbands. Each high-frequency subband is further divided into multiple directions (typically six directions: ±15°, ±45°, and ±75°), resulting in multi-scale and multi-directional characteristic subbands. For plastic packaging materials in nuclear facility environments, the dual-tree complex wavelet transform is highly effective in detecting microcracks because it can capture edge and texture features in different directions, particularly those fine linear cracks caused by radiation.

[0051] Extracting features from multi-scale, multi-directional feature subbands is crucial for describing material surface structure. First, local second-order matrix features are extracted. By calculating the covariance matrix of the local region of each feature subband, its eigenvalues ​​and eigenvectors are extracted to describe the directionality and anisotropy of the region. Gabor texture feature extraction uses a Gabor filter bank of varying scales and orientations to convolve the image, capturing texture information at varying frequencies and orientations. This approach is particularly well-suited for describing periodic textures and irregular microcracks on the surface of plastic packaging materials. Morphological feature extraction applies morphological operations such as opening and closing operations and skeleton extraction to the feature subbands, calculating topological properties such as regional connectivity and the number of branching points. This effectively describes the shape characteristics of defects such as bubbles and deformations. These three types of features are combined to form a structural feature vector, which comprehensively characterizes the surface structural properties of spacecraft electronic component packaging materials. Feature extraction is performed by converting the rectified image from RGB color space to HSV color space. The HSV color space decomposes color into three channels: hue, saturation, and value. This decomposition is more consistent with human color perception and is less sensitive to changes in lighting. The HSV color space is particularly well-suited for quantitative analysis of the characteristic discoloration of plastic encapsulating materials caused by gamma-ray radiation. During the extraction process, statistical features are calculated from three channels: the mean reflects the average level of the channel; the standard deviation describes the dispersion of the color distribution; the skewness measures the asymmetry of the color distribution; the kurtosis indicates the sharpness of the distribution; and the energy and entropy describe the concentration and disorder of the color distribution, respectively. These statistics form a color eigenvector, which accurately quantifies the characteristic discoloration of the material caused by radiation. After concatenating the structural eigenvector and the color eigenvector to form the initial eigenvector, the feature dimension is typically high and contains a large amount of redundant information. Dimensionality reduction is performed using principal component analysis (PCA) to retain the essential information. PCA calculates the covariance matrix of the eigenvectors, solves for their eigenvalues ​​and eigenvectors, and selects the first few eigenvectors corresponding to the largest eigenvalue as the projection basis. This achieves dimensionality reduction while retaining the main variation information of the data. The high-dimensional eigenvector is projected into a lower-dimensional space to obtain the reduced-dimensional eigenvector.

[0052] Linear Discriminant Analysis (LDA) is further applied to the reduced dimensionality vectors. Unlike PCA, LDA is a supervised dimensionality reduction method that considers class information. By maximizing the ratio of the between-class scatter matrix to the within-class scatter matrix, it finds the projection direction that best distinguishes different classes. In the context of defect detection for plastic packaging materials, LDA maximizes the separability of different defect types (microcracks, bubbles, deformation, and color difference) in the feature space, thereby obtaining an optimized eigenvector. A defect feature enhancement algorithm is then applied to the optimized eigenvector, using a nonlinear mapping function to enhance the distinction between defective and normal regions in the feature space. This nonlinear mapping function, typically a Sigmoid function, stretches samples near the feature space boundary, increasing the distance between defective and normal regions in the feature space. Based on this, a multiscale feature pyramid is constructed, retaining key features at different scales to capture information about defects of varying sizes. Ultimately, a multidimensional feature representation of data is obtained, encompassing typical microcracks, bubbles, deformation, and color difference characteristics found in nuclear facility environments.

[0053] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0054] Construct an AR-DCNN network with an encoder-decoder structure. The encoder contains multiple encoding blocks, and each encoding block has three parallel multi-scale convolution units to obtain a multi-scale feature map.

[0055] A radiation feature enhancement module is connected after the encoder, and the multi-scale feature map is processed through a residual connection block and a channel attention mechanism to obtain an enhanced feature representation. The channel attention mechanism is used to enhance the channel response of the plastic encapsulation material characteristics of spacecraft electronic components.

[0056] The enhanced feature representation is processed through multiple dilated convolutional layers of the decoder to form a multi-scale receptive field, obtaining feature parsing results at different scales. It is then upsampled by transposed convolution between decoding blocks and fused with the feature map of the corresponding encoding layer to obtain a fused feature map.

[0057] The fused feature map is input into the multi-task learning head, which includes three parallel branches: classification head, segmentation head, and scoring head, to obtain the initial defect prediction result;

[0058] The initial defect prediction results are optimized using a combined loss function, which includes weighted cross entropy loss for defect classification, Dice loss for segmentation, and mean square error loss for scoring, to obtain an optimized defect detection model.

[0059] The optimized defect detection model is used to process newly input plastic packaging material images in a nuclear facility environment, and output defect type identification and severity scoring results.

[0060] Specifically, based on the acquired multi-dimensional feature representation data, an encoder-decoder AR-DCNN (Anti-Radiation Deep Convolutional Neural Network) was constructed for defect detection and assessment. The AR-DCNN network is specifically designed for defect detection in plastic encapsulation materials under radiation environments and exhibits exceptional resistance to radiation noise. The encoder consists of five encoding blocks, each of which employs three parallel multi-scale convolutional units, using kernel sizes of 3×3, 5×5, and 7×7, respectively. This parallel design enables the network to simultaneously capture features at different scales. Each convolution operation is followed by a batch normalization layer and a LeakyReLU activation function. Batch normalization standardizes the input of each layer to accelerate training convergence, while LeakyReLU uses a slope of 0.2 on the negative axis to prevent neuron death. The number of convolution kernels is gradually increased from 64 in the first layer to 512 in the fifth layer. Max pooling is used between layers for downsampling, reducing the feature map size while preserving significant features, ultimately resulting in a multi-scale feature map.

[0061] The radiation feature enhancement module is connected after the encoder, which is the key component that distinguishes the AR-DCNN network from ordinary deep convolutional networks. This module contains a residual connection block and a channel attention mechanism. The residual connection block adopts a pre-activation design, that is, batch normalization and activation are performed before the convolution operation, which effectively alleviates the gradient vanishing problem of deep networks. The channel attention mechanism is specifically used to enhance the characteristic channel response of the plastic encapsulation material of spacecraft electronic components. Its working principle is to weight different feature channels according to their importance. First, the input feature map is subjected to global average pooling and maximum pooling to generate two channel descriptors respectively. These two descriptors are then processed by a shared multi-layer perceptron. Finally, the processed results are added and passed through a Sigmoid activation function to obtain weights between 0 and 1. These weights are multiplied by each channel of the original feature map to enhance the response strength of important feature channels and suppress irrelevant channels, thereby highlighting the defect characteristics unique to radiation environments and obtaining an enhanced feature representation.

[0062] The enhanced feature representation is processed through multiple dilated convolutional layers in the decoder. Dilated convolution, also known as dilated convolution, inserts holes in the convolution kernel to expand the receptive field while maintaining the number of parameters and computational complexity. The decoder consists of four decoding blocks, each of which uses 3×3 convolutional layers with different dilation rates (1, 2, 4, and 8), forming a multi-scale receptive field and capable of simultaneously detecting defect features of different sizes. Upsampling is performed between decoding blocks using transposed convolutions. The transposed convolution operation uses learned parameters to restore low-resolution feature maps to high-resolution feature maps, with a stride of 2, achieving a 2x upsampling. The upsampled feature maps are fused with the corresponding feature maps from the encoding layer via skip connections. The fusion method uses a weighted summation, with weights automatically learned through network training. Skip connections transfer high-resolution detail from the encoding stage directly to the decoding stage, compensating for the spatial detail lost during upsampling, ultimately resulting in a fused feature map.

[0063] The fused feature map is input into the multi-task learning head, the output of the AR-DCNN network. It consists of three parallel branches: the classification head, the segmentation head, and the scoring head. The classification head converts the feature vector into a probability distribution of defect types through a fully connected layer. It outputs confidence scores for five categories, including four major defects: microcracks, bubbles, deformation, and color difference, as well as normal areas. The segmentation head maps the feature map into a binary mask through a 1×1 convolution, identifying the precise location and shape of the defect area. The scoring head compresses the feature map into a one-dimensional vector through global average pooling, and then maps it into a severity score ranging from 0 to 10 through a fully connected layer. The outputs of these three branches together constitute the initial defect prediction result, providing comprehensive information on the defect type, location, and severity.

[0064] A combined loss function is applied to optimize the initial defect prediction results. The design of a combined loss function is key to multi-task learning. The defect classification task uses a weighted cross-entropy loss, with weights set to 3:2:2:2:1 for cracks: bubbles: deformation: color difference: normal, reflecting the varying importance of different defect types. The segmentation task uses a combination of Dice loss and weighted cross-entropy. Dice loss focuses on increasing the overlap of segmentation masks and is particularly effective for addressing class imbalance. The scoring task uses mean squared error loss, which directly measures the difference between the predicted and true scores. The losses of the three tasks are weighted and summed according to the weighted ratio of classification: segmentation: scoring = 0.3:0.5:0.2 to form the final combined loss. The network parameters are updated and the loss function is optimized through backpropagation, ultimately resulting in an optimized defect detection model.

[0065] When the optimized defect detection model is used to process newly input images of plastic packaging materials in a nuclear facility environment, the image first undergoes the same preprocessing steps as in the training phase. Then, the image is segmented into overlapping local regions using a sliding window approach and fed into the AR-DCNN network. Each window generates a set of predictions, which are then combined using an ensemble decision strategy (such as weighted voting) to produce the final output: defect type identification and severity score.

[0066] In a specific embodiment, the process of inputting the fused feature map into the multi-task learning head, including the three parallel branch steps of the classification head, the segmentation head, and the scoring head, may specifically include the following steps:

[0067] The fused feature map is input into the classification head and transformed through the fully connected layer to obtain the probability distribution of the defect types of the plastic packaging material. The defect types include microcracks, bubbles, deformation, color difference and normal areas that are unique to radiation environments.

[0068] Apply the 1×1 convolution operation in the segmentation head to the fused feature map to perform pixel-level feature mapping and obtain a binary defect area segmentation mask;

[0069] The fused feature map is input into the scoring head after global average pooling and mapped through the fully connected layer to obtain the defect severity score;

[0070] Perform the argmax operation on the defect type probability distribution, extract the defect type label corresponding to the highest probability, and obtain the defect type prediction result;

[0071] Apply morphological post-processing to the defect area segmentation mask, including opening and closing operations and area filtering, to obtain a refined defect area boundary;

[0072] The defect type prediction results, refined defect area boundaries and defect severity scores are integrated to form a complete initial defect prediction result for the plastic packaging materials of spacecraft electronic components.

[0073] Specifically, the fused feature map is input into the classification head for feature transformation. The classification head consists of multiple fully connected layers, each of which incorporates linear transformations and nonlinear activation functions. Specifically, the fused feature map is first compressed using global average pooling to convert the spatial dimensions into a one-dimensional feature vector. It then passes through two fully connected layers, the first using the ReLU activation function and the second without, directly outputting the raw scores. Finally, the raw scores are converted to a normalized probability distribution using the Softmax function to obtain a probability distribution of defect types for the encapsulation material. These defect types include microcracks, bubbles, deformation, color aberration, and normal areas, which are unique to radiation environments. Each type corresponds to a probability value, and the sum of all probabilities is 1. Applying the segmentation head's 1×1 convolution operation to the fused feature map is the process of performing pixel-level feature mapping. The 1×1 convolution operation refers to a convolution layer with a kernel size of 1×1. Its purpose is to change the number of channels and enable information exchange between channels without changing the spatial size of the feature map. In the segmentation head, a 1×1 convolution maps the number of channels in the fused feature map from the original multi-channel (such as 256 or 512) to the number of target categories (usually set to 2 in radiation environments, representing defective and non-defective areas). The output at each pixel position is the probability value of the pixel belonging to the defective area, ranging from 0 to 1. The probability map is thresholded (usually set to 0.5), and pixels greater than the threshold are marked as 1 (defective area) and pixels less than the threshold are marked as 0 (non-defective area), resulting in a binary defect area segmentation mask.

[0074] The process of evaluating defect severity is to pass the fused feature map through global average pooling and then input it into the scoring head. Global average pooling is a special pooling operation that calculates the global average for each feature channel, reducing the entire feature map to a one-dimensional vector, where each element represents the average activation intensity of a channel. This operation not only significantly reduces the number of parameters but also preserves the global information in the feature map. The scoring head receives this one-dimensional vector and maps it to a scalar value representing the defect severity score through one or more fully connected layers. The score range is typically set to 0-10, where 0 indicates no defect and 10 indicates the most severe defect.

[0075] Performing the argmax operation on the defect type probability distribution extracts the final predicted category from the probability distribution. The argmax operation finds the index with the highest probability value in the probability distribution, indicating the most likely defect type. For example, if the probability distribution is [0.05, 0.82, 0.06, 0.04, 0.03] and the corresponding defect types are [microcrack, bubble, deformation, color difference, normal], the argmax operation returns index 1, indicating that the predicted defect type is bubble. This hard decision method converts the probability distribution into a clear defect type label, resulting in a defect type prediction result.

[0076] Applying morphological post-processing to the defect region segmentation mask is a key step in improving the quality of segmentation results. Morphological post-processing includes opening and closing operations and area filtering. The opening operation is a combination of erosion followed by dilation, which is used to remove small noise and burrs; the closing operation is a combination of dilation followed by erosion, which is used to fill small holes and disconnected areas. In actual processing, the closing operation is first applied to fill small holes in the defect region, and then the opening operation is applied to remove noise and burrs at the edges. Area filtering removes small connected areas, which are often false positives caused by noise. Through these morphological operations, a refined defect region boundary with smooth boundaries and internal coherence is obtained.

[0077] The final step in forming a complete defect prediction is to fuse the defect type prediction results, the refined defect region boundaries, and the defect severity score. The fusion process first assigns a unique identifier to each detected defect region and then associates the defect type label and severity score with the corresponding defect region. For cases with multiple isolated defect regions, connected component analysis is used to divide the segmentation mask into independent defect instances, each containing attribute information such as location, area, type, and severity. The resulting complete initial defect prediction for the plastic encapsulation material of spacecraft electronic components contains all key defect information, providing a foundation for subsequent performance evaluation and life prediction.

[0078] For example, plastic encapsulation materials used in aerospace electronic components require defect detection after exposure to high-radiation environments. The 128×128×256 fused feature map obtained from the AR-DCNN network is fed into the classification head of the multi-task learning head. Global average pooling is used to generate a 256-dimensional feature vector. This is then passed through two fully connected layers (first layer 256×128, second layer 128×5) to produce a five-dimensional vector [-2.1, 4.3, 0.9, -0.5, -3.2] representing the raw scores for the five defect categories. Applying the Softmax function yields a probability distribution of [0.01, 0.87, 0.08, 0.03, 0.01], indicating an 87% probability of a bubble defect in this region. Simultaneously, the segmentation head applies a 1×1 convolution to the fused feature map, mapping the 256 channels to 2 channels. A sigmoid function is then applied to generate a defect probability map for each pixel. After thresholding (threshold 0.5), the initial binary mask is obtained, in which the continuous white area indicates the potential defect location. A structuring element with a radius of 2 is applied for closing operation to fill several small holes inside the defect; then a structuring element with a radius of 1 is applied for opening operation to smooth the defect edge; finally, the connected areas with an area less than 10 pixels are removed to obtain a refined defect area boundary. The scoring head receives the feature vector after global average pooling and maps it to a defect severity score of 7.8 through the fully connected layer, indicating that this is a relatively serious defect. Through connected component analysis, the segmentation mask is divided into 3 independent defect instances, each located at a different position of the material. After each instance is associated with its predicted type and score value, a complete initial defect prediction result is formed, which successfully detects and evaluates the radiation-induced defects on the plastic packaging material of spacecraft electronic components.

[0079] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0080] Based on the defect type identification and severity scoring results, a density-weighted Voronoi diagram is constructed to associate the detected defects with the surrounding areas to obtain a radiation damage pattern map.

[0081] The kernel density estimation algorithm is applied to the radiation damage pattern map, and the Gaussian kernel function is used to calculate the damage distribution density function of the material surface to obtain the damage distribution characteristics of the plastic packaging material surface.

[0082] Based on the surface damage distribution characteristics of the plastic packaging material, support vector regression was used to analyze the correlation between different types of defects and radiation exposure, distinguishing defects caused by radiation from those caused by non-radiation factors, and obtaining a quantitative index of radiation damage.

[0083] The radiation damage quantitative indicators are input into the random forest algorithm to build a plastic encapsulation material performance degradation assessment model, which outputs the predicted values ​​of key performance parameters of plastic encapsulation materials used in nuclear facilities, including electrical performance, mechanical strength, and thermal stability indicators;

[0084] The Weibull distribution is applied to the predicted values ​​of key performance parameters to establish a functional relationship between the performance degradation rate and time, and the material performance degradation curve is obtained. The Weibull distribution is particularly suitable for describing the failure mode of plastic packaging materials in a gamma-ray radiation environment.

[0085] The estimated time when the performance drops to a specified percentage of the initial value is determined based on the material performance degradation curve, and multi-level safety thresholds are set to obtain safety threshold warning information for the use of nuclear facility plastic sealing materials.

[0086] Specifically, the density-weighted Voronoi diagram is a spatial segmentation method that divides a plane into multiple regions, each corresponding to a seed point (in this method, the detected defect point). Unlike the traditional Voronoi diagram, the density-weighted Voronoi diagram takes into account the weight of the seed point. The larger the weight of the seed point, the larger the region corresponding to the seed point. In the actual construction process, each detected defect point is first used as a seed point, and the weight calculation formula is W = S × (0.7A + 0.3G), where S is the defect severity score (range 0-10), A is the normalized defect area (a value between 0-1, obtained by dividing the defect area by the maximum possible area), and G is the normalized gradient value (a value between 0-1, indicating the clarity of the defect edge). The Fortune algorithm is used to generate the Voronoi diagram. This algorithm is based on the scan line principle, scanning the seed points on the plane from left to right, and gradually constructing the Voronoi boundary. For each pair of adjacent seed points, the position of the Voronoi boundary is calculated based on their position and weight to form a complete Voronoi diagram. The resulting density-weighted Voronoi diagram intuitively displays the distribution pattern of radiation damage on the material surface, and each defect point and its affected area form a radiation damage pattern map. Applying the kernel density estimation algorithm to the radiation damage pattern map is an effective method to quantify the damage distribution density. Kernel density estimation is a non-parametric density estimation technique that places a kernel function on each data point and then sums all kernel functions to obtain a smooth density distribution. In this method, the Gaussian kernel function is used for kernel density estimation, which has good smoothness and mathematical properties. For each position point on the material surface, the damage density value of the position is calculated based on its distance from each defect point and the defect weight. The bandwidth parameter h of the kernel density estimation is automatically determined by the Silverman rule: h = 0.9×min(σ,R / 1.34)×n (-1 / 5), where σ is the standard deviation of the seed points, R is the interquartile range, and n is the number of seed points. By performing kernel density estimation on the entire material surface, a continuous damage distribution density function is obtained, forming a damage distribution characteristic for the plastic encapsulation material surface. This distribution characteristic can intuitively identify areas with severe and less severe damage, facilitating subsequent correlation analysis.

[0087] Using support vector regression (SVR) to analyze the correlation between different types of defects and radiation exposure based on the surface damage distribution characteristics of plastic encapsulation materials is key to identifying defect causes. SVR is a machine learning method that maps data into a high-dimensional feature space to find a hyperplane that minimizes prediction error. In this method, the input features include defect geometry, texture, and spatial distribution characteristics (a total of 58 dimensions), and the output is the correlation coefficient (between 0 and 1) between the defect and radiation exposure. SVR uses the RBF kernel function. The kernel parameter γ determines the kernel width, the penalty coefficient C controls the balance between model complexity and fitting error, and ε defines the range of error excluded. By analyzing the relationship between defect characteristics and historical radiation exposure data, the SVR model can learn patterns of radiation-specific defects. Defects with a correlation coefficient greater than 0.7 are classified as radiation-induced, those less than 0.3 as non-radiation-induced, and those between these two factors as mixed. This analysis yields quantitative radiation damage metrics, including the total area percentage of radiation-induced defects, the average depth of radiation-induced defects, and the spatial distribution characteristics of radiation-induced defects.

[0088] Inputting the radiation damage quantification index into the random forest algorithm to construct a performance degradation assessment model for plastic packaging materials is a core step in predicting material performance. Random forest is an integrated learning method consisting of multiple decision trees. Each tree is trained independently, and the average or majority vote result is taken when predicting. During the construction process, the radiation damage quantification index is first normalized to unify the numerical range of each index. The standardized damage index vector is then paired with the material performance parameters in the historical experimental data to form a training data set. During random forest training, each decision tree is trained using a randomly sampled data subset (the sampling ratio is usually 70%) and a feature subset (the number of features is usually the square root of the total number of features), which enhances the generalization ability of the model. Each decision tree is constructed by recursive binary splitting, and the best splitting feature and threshold are selected at each node to maximize the purity of the child node. After the training is completed, for the newly input radiation damage quantitative indicators, the final prediction result is obtained by averaging the predicted values ​​of all decision trees, and the predicted values ​​of key performance parameters of plastic packaging materials for nuclear facilities are output, including electrical properties (insulation strength, volume resistivity), mechanical strength (tensile strength, bending strength, hardness) and thermal stability (heat deformation temperature, thermal expansion coefficient) indicators.

[0089] Fitting the predicted values ​​of key performance parameters with a Weibull distribution is an important step in establishing a performance degradation model. The Weibull distribution is a continuous probability distribution widely used in reliability engineering and life analysis. Its flexible mathematical form adapts to a wide range of failure data. The Weibull distribution is particularly well-suited for describing the failure modes of plastic encapsulation materials exposed to gamma-ray radiation, as radiation-induced material degradation typically follows a nonlinear pattern. During the fitting process, the performance parameter values ​​at different cumulative radiation doses are first normalized as a ratio relative to the initial value. Maximum likelihood estimation is then used to determine the shape and scale parameters of the Weibull distribution. The shape parameter describes the temporal trend of the failure rate, while the scale parameter is related to the material's characteristic lifespan. Using the fitted Weibull distribution parameters, a functional relationship between the performance degradation rate and the cumulative radiation dose (or time) is established, resulting in a material performance degradation curve. Degradation curves are developed for different performance parameters (such as dielectric strength and mechanical strength) to comprehensively evaluate changes in material performance under radiation exposure.

[0090] The final step in generating early warning information is to determine the estimated time when the performance drops to a specified percentage of the initial value based on the material performance degradation curve. The performance degradation curve describes how the material performance changes with the cumulative radiation dose. These curves can be used to predict the performance value of the material at any radiation dose in the future. For nuclear facility plastic sealing materials, multi-level safety thresholds are usually set, such as 90%, 80%, and 70% of the initial value as the third-level warning threshold. By solving the degradation curve equation, the cumulative radiation dose value when the performance reaches these thresholds is found, and then the corresponding time point is converted according to the radiation intensity (dose rate) of the facility. These time points constitute the safety threshold warning information for the use of plastic sealing materials, guiding nuclear facility operators in material replacement or maintenance plans.

[0091] In a specific embodiment, the process of inputting the radiation damage quantitative index into the random forest algorithm to construct a plastic packaging material performance degradation assessment model may specifically include the following steps:

[0092] The radiation damage quantitative indicators are normalized to unify the numerical range of each indicator and obtain a standardized damage indicator vector;

[0093] The standardized damage index vector is matched with samples in the historical experimental database to construct a training data set, and a sample set containing input features and output performance parameters is obtained;

[0094] Apply the random forest algorithm to the sample set to construct multiple decision trees. Each decision tree is trained using a randomly sampled feature subset to obtain a set of decision trees.

[0095] Perform ensemble learning on the decision tree set. For each performance indicator of the nuclear facility plastic packaging material, the average prediction value of all decision trees is used as the final prediction result to obtain the predicted value of electrical performance, including insulation strength and volume resistivity.

[0096] Predict mechanical strength indicators based on a decision tree set, calculate tensile strength, bending strength and hardness parameters, and obtain predicted mechanical properties;

[0097] The decision tree set is used to predict the thermal deformation temperature and thermal expansion coefficient to obtain the thermal stability prediction value, which is directly related to the heat resistance reliability of spacecraft electronic components in radiation environment.

[0098] Specifically, radiation damage quantification requires feature normalization. Radiation damage typically manifests as varying degrees of damage to plastic encapsulation materials exposed to radiation. This damage can be quantified using a number of physical or chemical performance indicators, such as dielectric strength, volume resistivity, heat deformation temperature, and tensile strength. However, the numerical ranges of these indicators vary significantly, and their direct use can compromise the training effectiveness of subsequent machine learning models. Therefore, feature normalization is essential—converting the data for each indicator to a uniform numerical range so that the subsequent model can effectively learn the relationships between different features.

[0099] One common method of normalization is to subtract the minimum value from each feature data and then divide it by the range of the feature (the difference between the maximum and minimum values) to obtain a standardized data between 0 and 1. Through this process, all damage indicator data will be within the same numerical range, thus avoiding excessive or insufficient weights for certain features in subsequent machine learning models. The standardized damage indicator vector will be matched with samples in the historical experimental database. Historical experimental data refers to the performance data of various plastic encapsulation materials under different radiation environments measured through past experiments. These data include a series of physical parameters such as the dielectric strength, volume resistivity, tensile strength, and flexural strength of the materials under different radiation environments, which serve as the output parameters of the sample set.

[0100] By matching standardized damage indicators with samples from a historical experimental database, a training dataset is constructed, consisting of input features (standardized damage indicators) and output performance parameters (such as dielectric strength and volume resistivity). This training dataset is used to train machine learning models, specifically the random forest algorithm. The random forest algorithm performs classification or regression tasks by constructing multiple decision trees. Each decision tree is constructed by randomly sampling the training data and selecting a subset of features. This approach effectively avoids overfitting and improves the model's generalization capabilities.

[0101] Random forests work by using multiple decision trees to predict input data. The predictions from each tree are then averaged or voted to arrive at the final prediction. For example, suppose multiple decision trees are trained on a training set. Each tree predicts a material's electrical properties (such as dielectric strength and volume resistivity) based on the input radiation damage normalization indicator vector. The final prediction is the average of all the tree predictions.

[0102] For electrical property predictions, such as dielectric strength and volume resistivity, the final prediction is obtained by averaging the outputs of each decision tree. In addition to electrical properties, random forests can also be used to predict parameters such as mechanical properties and thermal stability. Mechanical property predictions, including tensile strength, flexural strength, and hardness, are also performed using an ensemble learning approach using a set of decision trees. For example, if multiple decision trees are trained to predict tensile strength (e.g., each tree predicts 500 MPa, 510 MPa, 495 MPa, etc.), the final tensile strength prediction is the average of these values. For thermal stability, the ensemble of decision trees is also used to predict heat deformation temperature and thermal expansion coefficient. Thermal stability directly impacts the heat resistance reliability of plastic encapsulation materials used in spacecraft electronic components, making this prediction crucial. For example, if multiple decision trees predict heat deformation temperature as 300°C, 305°C, 290°C, 295°C, etc., the final prediction is also the average of these values.

[0103] The above describes the method for detecting defects of plastic packaging materials in a radiation environment based on machine vision in the embodiment of the present application. The following describes the system for detecting defects of plastic packaging materials in a radiation environment based on machine vision in the embodiment of the present application. Figure 2 In one embodiment of the present application, a system for detecting defects in plastic packaging materials in a radiation environment based on machine vision includes:

[0104] The enhancement module is used to perform image acquisition and adaptive contrast enhancement processing on the surface of the plastic packaging material of the spacecraft electronic components under a gamma-ray environment through a lead-tungsten alloy shielded high radiation tolerance image sensor to obtain a radiation-resistant enhanced image;

[0105] An extraction module is used to perform dual-tree complex wavelet transform feature decomposition and HSV color space defect feature extraction on the anti-radiation enhanced image to obtain multi-dimensional feature representation data including microcracks, bubbles, deformation, and color difference features;

[0106] An input module is used to input the multi-dimensional feature representation data into the AR-DCNN network with residual connection and attention mechanism to perform void convolution multi-task learning to obtain defect type recognition and severity scoring results;

[0107] The prediction module is used to predict the radiation resistance life of plastic packaging materials for nuclear facilities based on the defect type identification and severity scoring results through Weibull distribution fitting and random forest algorithm, and obtain the material performance degradation curve and safety threshold warning.

[0108] Through the collaborative efforts of the various components mentioned above, in terms of the standardized processing of radiation damage quantitative indicators, this solution uses feature normalization processing to enable comparison of various indicators within a unified numerical range, providing a reliable data basis for subsequent intelligent prediction. The standardized damage indicator vector can eliminate the influence between different dimensions, making the data more comparable and consistent, and providing effective input data for the training of machine learning models. Secondly, a training set is constructed based on samples in the historical experimental database. By matching the standardized damage indicator vector with the samples, a comprehensive prediction of material properties is achieved. The random forest algorithm is used to perform ensemble learning through multiple decision trees. Under diversified data input, it can effectively avoid overfitting and improve the accuracy and robustness of the prediction. The advantage of the random forest algorithm is that it can process high-dimensional data and has a strong modeling ability for complex relationships between features. Therefore, it can more accurately predict multiple performance indicators such as the electrical properties, mechanical strength, and thermal stability of plastic packaging materials. In particular, for electrical performance prediction, an ensemble learning approach based on a set of decision trees can use the average prediction value of all decision trees as the final result, thereby improving the prediction accuracy of dielectric strength and volume resistivity. This is crucial for the long-term reliability of electronic components in spacecraft and high-radiation environments. Similarly, for mechanical performance and thermal stability prediction, the decision tree model effectively combines quantitative radiation damage indicators to provide accurate predictions of mechanical properties such as tensile strength, flexural strength, and hardness, as well as thermal stability indicators such as heat deformation temperature and thermal expansion coefficient. These indicators play a decisive role in the thermal reliability of spacecraft electronic components in radiation environments. Accurate predictions provide theoretical support and data basis for material optimization and application. This solution incorporates artificial intelligence algorithms, particularly the random forest model, which significantly improves the efficiency of data processing and model prediction. Compared to traditional manual testing and calculation methods, random forests, by processing multiple decision trees in parallel, can rapidly extract underlying patterns and patterns from complex data, avoiding the subjectivity and errors of manual evaluation. More importantly, as the number of samples increases, random forests gradually improve prediction accuracy, requiring minimal human intervention and demonstrating a high level of automation and intelligence. This feature is particularly important for spacecraft electronic components and related fields that require large-scale data processing. It can quickly respond to new experimental data and achieve real-time and accurate damage assessment.

[0109] above Figure 2From the perspective of modular functional entities, the plastic packaging material defect detection system in a radiation environment based on machine vision in an embodiment of the present invention is described in detail. Below, from the perspective of hardware processing, the plastic packaging material defect detection equipment in a radiation environment based on machine vision in an embodiment of the present invention is described in detail.

[0110] Figure 3 FIG3 is a schematic diagram of a structure of a device for detecting defects in molding material under a radiation environment based on machine vision, provided by an embodiment of the present invention. The device 300 may vary significantly depending on configuration or performance, and may include one or more processors (central processing units, CPUs) 310 (e.g., one or more processors), a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage medium 330 may be either transient or persistent storage. The program stored in the storage medium 330 may include one or more modules (not shown), each of which may include a series of instructions for operating on the device 300 for detecting defects in molding material under a radiation environment based on machine vision. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, executing the series of instructions stored in the storage medium 330 on the device 300 for detecting defects in molding material under a radiation environment based on machine vision, thereby implementing the steps of the aforementioned method for detecting defects in molding material under a radiation environment based on machine vision.

[0111] The device 300 for detecting defects in molding materials under radiation environment based on machine vision may further include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be understood by those skilled in the art that Figure 3 The structure of the device for detecting defects in plastic packaging materials in a radiation environment based on machine vision shown does not constitute a limitation on the device for detecting defects in plastic packaging materials in a radiation environment based on machine vision provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0112] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions. When the instructions are run on a computer, the computer executes the steps of the method for detecting defects in plastic packaging materials in a radiation environment based on machine vision.

[0113] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0114] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a plastic packaging material defect detection device (which can be a personal computer, server, or network device, etc.) based on machine vision in a radiation environment to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0115] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting defects in plastic packaging materials in a radiation environment based on machine vision, characterized in that: The method comprises: The surface of the plastic packaging material of the spacecraft electronic components is shielded by a lead-tungsten alloy and highly radiation-tolerant image sensor. Image acquisition and adaptive contrast enhancement processing are performed under a gamma-ray environment to obtain a radiation-resistant enhanced image. Performing dual-tree complex wavelet transform feature decomposition and HSV color space defect feature extraction on the radiation-resistant enhanced image to obtain multi-dimensional feature representation data including microcracks, bubbles, deformation, and color difference features; The multi-dimensional feature representation data is input into the AR-DCNN network with residual connection and attention mechanism for void convolution multi-task learning to obtain defect type recognition and severity scoring results, including: constructing an AR-DCNN network with an encoder-decoder structure, the encoder includes multiple encoding blocks, each encoding block is provided with three parallel multi-scale convolution units to obtain a multi-scale feature map; connecting a radiation feature enhancement module after the encoder, processing the multi-scale feature map through the residual connection block and the channel attention mechanism to obtain an enhanced feature representation, wherein the channel attention mechanism is used to enhance the characteristic channel response of the plastic packaging material of the spacecraft electronic component; processing the enhanced feature representation through multiple void convolution layers of the decoder to form Multi-scale receptive field, obtains feature analysis results at different scales, and upsamples them between decoding blocks through transposed convolution, and fuses them with the feature map of the corresponding encoding layer to obtain a fused feature map; the fused feature map is input into the multi-task learning head, which includes three parallel branches: classification head, segmentation head, and scoring head, to obtain an initial defect prediction result; the initial defect prediction result is optimized by applying a combined loss function, which includes weighted cross entropy loss for defect classification, Dice loss for segmentation, and mean square error loss for scoring, to obtain an optimized defect detection model; the optimized defect detection model is used to process newly input plastic packaging material images in a nuclear facility environment, and output defect type identification and severity scoring results; Based on the defect type identification and severity scoring results, the radiation resistance life of plastic packaging materials for nuclear facilities is predicted through Weibull distribution fitting and random forest algorithm, and the material performance degradation curve and safety threshold warning are obtained.

2. The method for detecting defects of plastic packaging materials in a radiation environment based on machine vision according to claim 1, characterized in that: The method of performing image acquisition and adaptive contrast enhancement processing on the surface of the plastic packaging material of the spacecraft electronic component by using a lead-tungsten alloy shielded high radiation tolerance image sensor in a gamma ray environment to obtain a radiation-resistant enhanced image includes: A lead-tungsten alloy composite shielding layer of a specified thickness is provided on the exterior of the image sensor to obtain a radiation-resistant sensor with a high shielding ratio. A radiation-resistant reinforcement scheme with a multiple redundancy design is implemented on the internal circuit of the radiation-resistant sensor to obtain a circuit system that maintains normal function in a high-energy radiation environment. The sensor's radiation exposure level is monitored in real time through the parameter automatic correction unit. When the radiation intensity exceeds the preset threshold, the camera's sensitivity parameters and exposure time are automatically adjusted to obtain image acquisition parameters suitable for the current radiation environment. Acquire a surface image of the plastic packaging material according to the image acquisition parameters to obtain an image of the material under the original radiation environment; Decomposing the material image in the original radiation environment into multiple frequency sub-bands by wavelet decomposition, and applying adaptive threshold denoising related to the current radiation intensity to the high frequency sub-bands to obtain a denoised image; Applying an adaptive contrast enhancement algorithm with a fixed pixel window size and a set overlap rate to the denoised image for local enhancement, wherein the enhancement coefficient is dynamically determined according to the local entropy value of the image, to obtain a preliminary enhanced image; According to the scattering, drift, and thermal noise radiation interference feature patterns contained in the interference feature library, the interference type in the preliminary enhanced image is identified by pattern matching to obtain image interference features; According to the image interference characteristics, the corresponding compensation strategy is applied to correct the interference of the preliminary enhanced image to obtain the final radiation-resistant enhanced image, and the current radiation type, radiation intensity, and cumulative dose are recorded to form radiation environment parameter data.

3. The method for detecting defects of plastic packaging materials in a radiation environment based on machine vision according to claim 1, characterized in that: The step of performing dual-tree complex wavelet transform feature decomposition and HSV color space defect feature extraction on the radiation-resistant enhanced image to obtain multi-dimensional feature representation data including microcracks, bubbles, deformation, and color difference features includes: Applying a geometric deformation correction algorithm to the radiation-resistant enhanced image, eliminating image distortion caused by radiation through a bicubic B-spline transformation model with multiple control points, and obtaining a corrected image with sub-pixel accuracy; Applying a dual-tree complex wavelet transform to the corrected image to decompose the image into multiple decomposition scales, each scale containing multiple directional subbands, thereby obtaining multi-scale and multi-directional feature subbands. The dual-tree complex wavelet transform is a surface texture analysis method for plastic packaging materials unique to nuclear radiation environments and has a significant detection effect on radiation microcracks. Extracting local second-order matrix features, Gabor texture features, and morphological features from the multi-scale and multi-directional feature subbands to obtain a structural feature vector of the spacecraft electronic component packaging material; Converting the corrected image to the HSV color space, extracting statistical features of the three channels of hue, saturation, and lightness, including mean, standard deviation, skewness, kurtosis, energy, and entropy, to obtain a color feature vector, wherein the HSV color space feature extraction quantifies the characteristic discoloration of the plastic packaging material caused by gamma-ray radiation; The structural feature vector and the color feature vector are connected in series to form an initial feature vector, and the feature dimension is reduced to a lower dimension by principal component analysis, retaining the main information, to obtain a reduced-dimensional feature vector; Further reducing the dimension of the reduced feature vector by linear discriminant analysis while maximizing the separability between different defect categories to obtain an optimized feature vector; A defect feature enhancement algorithm is applied to the optimized feature vector, and the difference between the defect area and the normal area in the feature space is enhanced through a nonlinear mapping function. A multi-scale feature pyramid is constructed to obtain multi-dimensional feature representation data containing typical microcracks, bubbles, deformation, and color difference features in a nuclear facility environment.

4. The method for detecting defects of plastic packaging materials in a radiation environment based on machine vision according to claim 1, characterized in that: The fused feature map is input into a multi-task learning head, including three parallel branches: a classification head, a segmentation head, and a scoring head, to obtain an initial defect prediction result, including: The fused feature map is input into the classification head and feature transformed through the fully connected layer to obtain the probability distribution of the defect types of the plastic packaging material. The defect types include microcracks, bubbles, deformation, color difference and normal areas that are unique to radiation environments; Applying a 1×1 convolution operation in the segmentation head to the fused feature map to perform pixel-level feature mapping to obtain a binary defect area segmentation mask; The fused feature map is input into the scoring head after global average pooling, and mapped through the fully connected layer to obtain the defect severity score value; Performing an argmax operation on the defect type probability distribution, extracting the defect type label corresponding to the highest probability, and obtaining a defect type prediction result; Applying morphological post-processing to the defect area segmentation mask, including opening and closing operations and area filtering, to obtain a refined defect area boundary; The defect type prediction results, refined defect area boundaries and defect severity score values ​​are integrated to form a complete initial defect prediction result for the plastic packaging material of the spacecraft electronic components.

5. The method for detecting defects of plastic packaging materials in a radiation environment based on machine vision according to claim 1, characterized in that: The radiation resistance life of plastic packaging materials for nuclear facilities is predicted based on the defect type identification and severity scoring results through Weibull distribution fitting and random forest algorithm to obtain material performance degradation curve and safety threshold warning, including: Constructing a density-weighted Voronoi diagram based on the defect type identification and severity scoring results, correlating the detected defects with the surrounding areas, and obtaining a radiation damage pattern map; Applying a kernel density estimation algorithm to the radiation damage pattern map, using a Gaussian kernel function to calculate the damage distribution density function of the material surface, and obtaining the damage distribution characteristics of the plastic packaging material surface; Based on the surface damage distribution characteristics of the plastic packaging material, support vector regression is used to analyze the correlation between different types of defects and radiation exposure, distinguish defects caused by radiation from defects caused by non-radiation factors, and obtain a quantitative index of radiation damage; Inputting the radiation damage quantitative index into a random forest algorithm to construct a plastic encapsulation material performance degradation assessment model, and outputting predicted values ​​of key performance parameters of plastic encapsulation materials for nuclear facilities, including electrical performance, mechanical strength, and thermal stability indicators; Applying a Weibull distribution to fit the predicted values ​​of the key performance parameters, establishing a functional relationship between the performance degradation rate and time, and obtaining a material performance degradation curve, wherein the Weibull distribution is particularly suitable for describing the failure mode of plastic packaging materials under a gamma-ray radiation environment; The estimated time when the performance drops to a specified percentage of the initial value is determined based on the material performance degradation curve, and multi-level safety thresholds are set to obtain safety threshold warning information for the use of nuclear facility plastic packaging materials.

6. The method for detecting defects of plastic packaging materials in a radiation environment based on machine vision according to claim 5, characterized in that: The radiation damage quantitative index is input into the random forest algorithm to construct a plastic packaging material performance degradation assessment model, and the key performance parameter prediction values ​​of the plastic packaging material for nuclear facilities are output, including electrical performance, mechanical strength and thermal stability indicators, including: Performing feature normalization processing on the radiation damage quantitative index to unify the numerical range of each index and obtain a standardized damage index vector; Matching the standardized damage index vector with samples in a historical experiment database to construct a training data set, thereby obtaining a sample set including input features and output performance parameters; Applying a random forest algorithm to the sample set to construct multiple decision trees, each decision tree is trained using a randomly sampled feature subset to obtain a set of decision trees; Performing ensemble learning on the decision tree set, and for each performance indicator of the nuclear facility plastic packaging material, using the average prediction value of all decision trees as the final prediction result to obtain the electrical performance prediction value, including insulation strength and volume resistivity; Predicting mechanical strength indicators based on the decision tree set, calculating tensile strength, bending strength and hardness parameters, and obtaining mechanical property prediction values; The decision tree set is used to predict the thermal deformation temperature and the thermal expansion coefficient to obtain a thermal stability prediction value, wherein the thermal stability prediction value is directly related to the heat resistance reliability of spacecraft electronic components in a radiation environment.

7. A system for detecting defects in plastic packaging materials in a radiation environment based on machine vision, characterized in that: The method for detecting defects of plastic packaging materials in a radiation environment based on machine vision according to any one of claims 1 to 6 is configured to implement the method, wherein the system for detecting defects of plastic packaging materials in a radiation environment based on machine vision comprises: The enhancement module is used to perform image acquisition and adaptive contrast enhancement processing on the surface of the plastic packaging material of the spacecraft electronic components under a gamma-ray environment through a lead-tungsten alloy shielded high radiation tolerance image sensor to obtain a radiation-resistant enhanced image; An extraction module is used to perform dual-tree complex wavelet transform feature decomposition and HSV color space defect feature extraction on the anti-radiation enhanced image to obtain multi-dimensional feature representation data including microcracks, bubbles, deformation, and color difference features; The input module is used to input the multi-dimensional feature representation data into the AR-DCNN network with residual connection and attention mechanism for void convolution multi-task learning to obtain defect type identification and severity scoring results, including: constructing an AR-DCNN network with an encoder-decoder structure, wherein the encoder includes multiple encoding blocks, each encoding block is provided with three parallel multi-scale convolution units to obtain a multi-scale feature map; connecting a radiation feature enhancement module after the encoder, processing the multi-scale feature map through the residual connection block and the channel attention mechanism to obtain an enhanced feature representation, wherein the channel attention mechanism is used to enhance the characteristic channel response of the plastic packaging material of the spacecraft electronic component; processing the enhanced feature representation through multiple void convolution layers of the decoder , forming a multi-scale receptive field, obtaining feature analysis results at different scales, and upsampling between decoding blocks through transposed convolution, and fusing with the feature map of the corresponding encoding layer to obtain a fused feature map; inputting the fused feature map into the multi-task learning head, including three parallel branches of the classification head, segmentation head and scoring head, to obtain an initial defect prediction result; applying a combined loss function to the initial defect prediction result for optimization, the combined loss function includes the weighted cross entropy loss for defect classification, the Dice loss for segmentation and the mean square error loss for scoring, to obtain an optimized defect detection model; using the optimized defect detection model to process the newly input plastic packaging material image in the nuclear facility environment, and output the defect type identification and severity scoring results; The prediction module is used to predict the radiation resistance life of plastic packaging materials for nuclear facilities based on the defect type identification and severity scoring results through Weibull distribution fitting and random forest algorithm, and obtain the material performance degradation curve and safety threshold warning.

8. A device for detecting defects in plastic packaging materials in a radiation environment based on machine vision, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the method for detecting defects of plastic packaging materials in a radiation environment based on machine vision according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is enabled to execute the method for detecting defects of plastic packaging materials in a radiation environment based on machine vision according to any one of claims 1 to 6.

Images