Internal Defect Detection Method, System and Storage Medium for Automobile Injection Molding Parts

Abstract

The present invention discloses a method, system and storage medium for detecting internal defects of automotive injection molded parts, which relates to the technical field of defect detection of injection molded parts and includes: scanning and imaging the automotive injection molded parts to be detected, identifying and marking abnormal areas, and obtaining N preliminary injection molded part abnormal areas; activating a secondary defect detection device to collect N multi-source detection images of the N preliminary injection molded part abnormal areas; performing registration alignment and multi-scale fusion to obtain N region multi-source fusion images, performing defect classification and positioning to determine an injection molded part region defect feature set; establishing a finite element model of the injection molded part to generate an internal defect detection result of the injection molded part. The present invention solves the technical problem in the prior art that it is difficult to comprehensively detect internal defects of automotive injection molded parts, resulting in insufficient detection accuracy and reliability, and achieves the comprehensive detection of internal defects of automotive injection molded parts and the evaluation of the impact of defects, improving the technical effects of the detection accuracy and reliability of internal defects of automotive injection molded parts.

Description

Technical Field

[0001] The present invention relates to the technical field of injection molded part defect detection, and particularly to an internal defect detection method, system and storage medium for automotive injection molded parts. Background Art

[0002] In the field of automotive manufacturing, injection molded parts are widely used as key components. However, during the production process of injection molded parts, due to various factors such as the quality of raw materials and fluctuations in injection molding process parameters, internal defects such as cracks, holes, and bubbles are likely to occur. Traditional injection molded part detection methods, which are based on single imaging technologies, such as only using visible light imaging, cannot penetrate the injection molded parts to obtain internal information; while only using X-ray imaging, although it can present the internal structure, it has low recognition of some defects with small density differences from surrounding materials. Moreover, existing technologies often lack the ability to comprehensively evaluate defects and cannot accurately judge the expansion risk of defects under the actual use conditions of injection molded parts and the degree of influence on their performance. These deficiencies make it difficult to efficiently and accurately detect internal defects of automotive injection molded parts, thereby affecting the overall quality and safety of automobiles.

[0003] There are technical problems in the prior art that it is difficult to comprehensively detect internal defects of automotive injection molded parts, resulting in insufficient detection accuracy and reliability. Summary of the Invention

[0004] This application provides an internal defect detection method, system and storage medium for automotive injection molded parts, aiming to solve the technical problems in the prior art that it is difficult to comprehensively detect internal defects of automotive injection molded parts, resulting in insufficient detection accuracy and reliability.

[0005] In view of the above problems, this application provides an internal defect detection method, system and storage medium for automotive injection molded parts.

[0006] In the first aspect of the embodiments of this application, an internal defect detection method for automotive injection molded parts is provided, and the method includes:

[0007] Use an infrared thermal imager to scan and image the automotive injection molded part to be detected, generate a thermal image of the surface of the injection molded part, identify and mark abnormal areas on the thermal image of the surface of the injection molded part, and obtain N preliminary abnormal areas of the injection molded part; activate the secondary defect detection device, and collect and obtain N multi-source detection images of the N preliminary abnormal areas of the injection molded part through the secondary defect detection device, where the N multi-source detection images include a visible light image of the injection molded part and an X-ray image of the injection molded part; register, align and multi-scale fuse the N multi-source detection images to obtain N region multi-source fusion images, and perform defect classification and positioning on the N region multi-source fusion images to determine the injection molded part region defect feature set; perform finite element simulation based on the injection molded part region defect feature set, establish a finite element model of the injection molded part, and perform solution simulation analysis and defect impact assessment on the finite element model of the injection molded part according to the use conditions of the injection molded part to generate the internal defect detection result of the injection molded part.

[0008] In the second aspect of the embodiments of the present application, an internal defect detection system for automotive injection molded parts is provided, and the system includes:

[0009] A preliminary abnormal area acquisition module of the injection molded part, which is used to use an infrared thermal imager to scan and image the automotive injection molded part to be detected, generate a thermal image of the surface of the injection molded part, identify and mark abnormal areas on the thermal image of the surface of the injection molded part, and obtain N preliminary abnormal areas of the injection molded part; a multi-source detection image acquisition module, which is used to activate the secondary defect detection device and collect and obtain N multi-source detection images of the N preliminary abnormal areas of the injection molded part through the secondary defect detection device, where the N multi-source detection images include a visible light image of the injection molded part and an X-ray image of the injection molded part; an injection molded part region defect feature set determination module, which is used to register, align and multi-scale fuse the N multi-source detection images to obtain N region multi-source fusion images, and perform defect classification and positioning on the N region multi-source fusion images to determine the injection molded part region defect feature set; a defect detection result generation module, which is used to perform finite element simulation based on the injection molded part region defect feature set, establish a finite element model of the injection molded part, and perform solution simulation analysis and defect impact assessment on the finite element model of the injection molded part according to the use conditions of the injection molded part to generate the internal defect detection result of the injection molded part.

[0010] In the third aspect of the embodiments of the present application, the present application provides a computer-readable storage medium storing a computer program, and the computer program is used to execute an internal defect detection method for automotive injection molded parts provided by the present application.

[0011] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0012] Use an infrared thermal imager to scan and image the automotive injection molded parts to be detected, generate a thermal image of the surface of the injection molded parts, identify and mark abnormal areas, and obtain N preliminary abnormal areas of the injection molded parts; activate the secondary defect detection device, and collect and obtain N multi-source detection images of the N preliminary abnormal areas of the injection molded parts through the secondary defect detection device; register, align and multi-scale fuse the N multi-source detection images to obtain N region multi-source fusion images, and classify and locate the defects of the N region multi-source fusion images to determine the region defect feature set of the injection molded parts; establish a finite element model of the injection molded parts, perform solution simulation analysis and defect impact assessment, and generate the internal defect detection result of the injection molded parts. It achieves the comprehensive detection of the internal defects of automotive injection molded parts and the assessment of the defect impact, and improves the technical effects of the accuracy and reliability of the internal defect detection of automotive injection molded parts. Brief Description of the Drawings

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0014] Figure 1 It is a schematic flowchart of a method for detecting internal defects of an automotive injection molded part provided by an embodiment of the present application.

[0015] Figure 2 It is a schematic structural diagram of a system for detecting internal defects of an automotive injection molded part provided by an embodiment of the present application.

[0016] Explanation of reference numerals: The module 10 for obtaining preliminary abnormal areas of the injection molded parts, the module 20 for obtaining multi-source detection images, the module 30 for determining the region defect feature set of the injection molded parts, and the module 40 for generating the defect detection result. Detailed Description of the Embodiments

[0017] The present application provides a method, system and storage medium for detecting internal defects of automotive injection molded parts, which are used to solve the technical problem that it is difficult to comprehensively detect the internal defects of automotive injection molded parts in the prior art, resulting in insufficient detection accuracy and reliability.

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts belong to the scope of protection of the present application.

[0019] Embodiment 1, as Figure 1As shown in the figure, the present application provides a method for detecting internal defects of automotive injection-molded parts, and the method includes:

[0020] Step S100: Use an infrared thermal imager to scan and image the automotive injection-molded part to be detected, generate a thermal imaging map of the surface of the injection-molded part, identify and mark abnormal regions on the thermal imaging map of the surface of the injection-molded part, and obtain N preliminary abnormal regions of the injection-molded part.

[0021] Specifically, the automotive injection-molded part to be detected is preliminarily screened by using an infrared thermal imager. The infrared thermal imager utilizes the thermal radiation characteristics of the object itself to scan the injection-molded part in all directions, and converts the temperature distribution on the surface of the injection-molded part into an intuitive thermal imaging map. This thermal imaging map is deeply analyzed. First, the distribution temperature value information of the thermal imaging map is extracted and converted into a grayscale map for subsequent processing. According to the noise characteristics of the thermal imaging map, a Gaussian filter is initialized, and the grayscale map is filtered to remove noise, obtaining a clear and usable infrared grayscale map of the surface of the injection-molded part. Based on the image contrast requirement, a suitable target pixel value threshold is set, and contrast enhancement processing is performed on it to make potential abnormal regions more obvious, generating a standard infrared grayscale map of the surface of the injection-molded part. Combining with the normal operating temperature threshold of the automotive injection-molded part to be detected, the initial abnormal region set is determined, seed points are selected to merge regions according to the neighborhood growth rule, and edge detection and refinement are performed using the Sobel operator to complete the identification and edge segmentation marking of the abnormal regions, thereby obtaining N preliminary abnormal regions of the injection-molded part, and these regions are the key objects of attention for subsequent detection.

[0022] Step S200: Activate the secondary defect detection device, and collect N multi-source detection images of the N preliminary abnormal regions of the injection-molded part through the secondary defect detection device. The N multi-source detection images include a visible light image of the injection-molded part and an X-ray image of the injection-molded part.

[0023] Specifically, after completing the preliminary screening of automotive injection molded parts and obtaining N preliminary abnormal regions of injection molded parts, start the secondary defect detection equipment, which integrates image acquisition technology. After the equipment is started, it quickly detects each preliminary abnormal region of the injection molded parts in sequence. It uses a high-resolution optical lens to capture the visible light image of the abnormal region of the injection molded part. Through the principle of light reflection and imaging, information such as the surface texture, color change, and shape contour of the abnormal region are clearly recorded to form a visible light image of the injection molded part, providing a basis for defect analysis from the external feature level. At the same time, the equipment emits X-rays to penetrate the abnormal region of the injection molded part. When the X-rays pass through the injection molded part, different gray-scale images are formed on the detector due to different absorption degrees of different materials and structures, and then an X-ray image of the injection molded part is generated, revealing the structural defect information hidden inside the injection molded part. For each preliminary abnormal region of the injection molded part, a set of multi-source detection images including the visible light image of the injection molded part and the X-ray image of the injection molded part is obtained, and a total of N groups of such multi-source detection images are obtained, laying a solid data foundation for subsequent accurate defect analysis.

[0024] Step S300: Register, align, and multi-scale fuse the N multi-source detection images to obtain N region multi-source fused images, and classify and locate the defects of the N region multi-source fused images to determine the injection molded part region defect feature set.

[0025] Specifically, processing is performed on N multi-source detection images obtained from a secondary defect detection device. To ensure that different modality images can be accurately corresponding in the same spatial coordinate system, these images are preprocessed for denoising in sequence to remove interference information generated during the acquisition process, and then feature points in the images are extracted to obtain N multi-source image feature point sets. Next, the similarity information between these feature point sets is calculated, and the matching feature point sets are found based on the similarity. Based on these matching feature points, spatial transformation and registration alignment operations are performed on the N multi-source detection images, so that the visible light image and the X-ray image are accurately corresponding in terms of spatial position, generating N multi-source registered images. Multi-scale processing is respectively performed on the N multi-source registered images. First, Gaussian blur processing and downsampling processing are performed on them, and N visible light image pyramids and N X-ray image pyramids are iteratively constructed. Multi-scale region division is performed on each layer of the pyramid to obtain N visible light multi-scale region sets and N X-ray multi-scale region sets. According to the requirements of the image fusion task, a region energy weighted fusion rule is obtained. According to this rule, based on the defined set of region energy metric indicators, the region energy of the N visible light multi-scale region sets and the N X-ray multi-scale region sets is respectively calculated to obtain N multi-scale visible light region energy sets and N multi-scale X-ray region energy sets. Then, the two multi-scale region sets are fused layer by layer with region weighting according to the region energy ratio to obtain N fused image pyramids, and then inverse transformation and image reconstruction are respectively performed on them, so as to obtain N region multi-source fused images. Finally, these N region multi-source fused images are analyzed using a pre-constructed automotive injection molding part defect classifier. This classifier is obtained through defect labeling and recognition training based on an automotive injection molding part defect image database. The classifier sequentially performs defect recognition and classification on the fused images to obtain an injection molding part defect feature type set, and at the same time, anchors are located for each defect feature respectively to obtain an injection molding part defect feature position set. By associating and integrating these two sets, the injection molding part region defect feature set is determined, clearly presenting key information such as the type and position of the injection molding part defects.

[0026] Step S400: Based on the injection molding part region defect feature set, perform finite element simulation, establish an injection molding part finite element model, perform solution simulation analysis and defect impact assessment on the injection molding part finite element model according to the service conditions of the injection molding part, and generate an internal defect detection result of the injection molding part.

[0027] Specifically, based on the regional defect feature set of the injection-molded part, finite element simulation is carried out. Using finite element analysis software, the geometric shape, material properties, and information such as the location and type of defects of the injection-molded part are digitally modeled to construct a finite element model of the injection-molded part. To make the simulation more conform to the actual situation, according to the actual working conditions of the injection-molded part, such as the pressure, temperature change, vibration frequency, etc., corresponding simulation load parameters are applied to the model. Then, a solution simulation analysis is carried out on the loaded model. During the solution process, physical quantities such as the stress, strain distribution, and displacement change of the model under various working conditions are calculated, especially focusing on the mechanical response of the defect area. After obtaining the simulation analysis results, based on the pre-set defect impact evaluation index set, a comprehensive evaluation of the internal defect expansion parameters of the injection-molded part is carried out. These indexes cover multiple aspects such as strength, stiffness, and fatigue life. By comprehensively considering the evaluation results, a detailed internal defect detection result of the injection-molded part is finally generated, which can clearly reflect the degree of influence of the defect on the performance of the injection-molded part, providing a strong basis for judging whether the injection-molded part is qualified, whether it can continue to be used, and subsequent repair or improvement measures.

[0028] In a possible implementation manner, step S100 further includes:

[0029] Step S110: Extract the distributed temperature value information of the thermal imaging map on the surface of the injection-molded part, convert the distributed temperature value information into a grayscale image, and obtain an infrared grayscale image of the surface of the injection-molded part.

[0030] Step S120: Initialize a Gaussian filter according to the noise characteristic information of the thermal imaging map on the surface of the injection-molded part, and use the Gaussian filter to filter and denoise the infrared grayscale image of the surface of the injection-molded part to obtain an available infrared grayscale image of the surface of the injection-molded part.

[0031] Step S130: Based on the application requirements of image contrast, set a target pixel value threshold, and map the available infrared grayscale image of the surface of the injection-molded part to the target pixel value threshold for contrast enhancement to obtain a standard infrared grayscale image of the surface of the injection-molded part.

[0032] Step S140: Identify and mark the abnormal areas and edges of the standard infrared grayscale image of the surface of the injection-molded part to obtain N preliminary abnormal areas of the injection-molded part.

[0033] Specifically, after obtaining the thermal imaging map of the injection molded part using an infrared thermal imager, a data extraction algorithm is used to traverse each pixel point in the thermal imaging map and accurately extract the distribution temperature value information corresponding to that point. This temperature value information reflects the heat distribution of different positions on the surface of the injection molded part and is an important basis for judging internal defects. Subsequently, an image conversion algorithm is adopted to convert the data presented in the form of temperature values into a grayscale image. In the grayscale image, different gray levels represent different temperature information. Through this conversion, the temperature data of the thermal imaging map is converted into a format that is more convenient for subsequent image processing and analysis, and finally an infrared grayscale image of the injection molded part surface is generated, providing a unified and easy-to-process data basis for subsequent operations such as image denoising, contrast enhancement, and defect recognition.

[0034] During the infrared thermal imaging acquisition process, various factors can cause noise in the image, and this noise will affect the accurate judgment of internal defects in the injection molded part. Therefore, first analyze the noise characteristics of the thermal imaging map of the injection molded part surface. By statistically analyzing information such as the distribution law and frequency characteristics of the noise in the image, determine the key parameters of the Gaussian filter, such as the standard deviation, etc., to complete the initialization of the Gaussian filter. The Gaussian filter is a linear smoothing filter, and its principle is based on the Gaussian function. After initialization, apply this filter to the infrared grayscale image of the injection molded part surface. During the filtering process, for each pixel point in the image, the filter will perform a weighted average calculation on the gray values of this pixel point and its neighboring pixel points according to the weights determined by the Gaussian function. In this way, effectively reduce the influence of noise on the pixel values, suppress the high-frequency noise in the image, and make the image smoother. After the filtering process, the image with most of the noise interference removed becomes the available infrared grayscale image of the injection molded part surface, providing clearer and more reliable image data for subsequent image analysis and defect recognition.

[0035] Process based on the application requirements of image contrast. According to experience or experimental data, set an appropriate target pixel value threshold. This threshold is a key parameter for contrast enhancement. Map the available infrared grayscale image of the injection molded part surface according to the set target pixel value threshold, adjust the distribution of pixel values in the image, and enhance the contrast of different regions in the image. After such processing, the originally possibly unobvious defect regions become clearer in the image, and a standard infrared grayscale image of the injection molded part surface is obtained, providing a clearer image basis for subsequent accurate defect recognition.

[0036] Use an image analysis algorithm to deeply analyze the infrared grayscale image of the surface of a standard injection molded part. By setting an appropriate temperature threshold range, which refers to the temperature distribution data of normal injection molded parts under specific conditions, to initially screen out areas that may be abnormal and form a starting set of abnormal areas. From this starting set, select multiple seed points according to certain rules. These seed points serve as the starting positions for subsequent region growth. According to the neighborhood growth rule, with the seed points as the center, examine the similarity between the grayscale values of its adjacent pixel points and the grayscale values of the seed points. If the similarity meets the set conditions, merge the adjacent pixel points into the current growing region and repeat this process continuously until there are no more eligible adjacent pixel points, thus obtaining a set of abnormal growth regions. To more accurately determine the boundary of the abnormal region, use the Sobel operator to perform edge detection on the set of abnormal growth regions. The Sobel operator highlights the edge information in the image by calculating the gradient changes of pixel points in the horizontal and vertical directions of the image. After the calculation, perform connection and refinement processing on the obtained edge information to remove some discontinuous noise points and redundant edges, thereby obtaining a clear and continuous set of edge contours of the abnormal region. Finally, based on these edge contours, perform precise edge segmentation and marking on the set of abnormal growth regions to clearly divide each independent abnormal region, and finally obtain N preliminary abnormal regions of the injection molded part.

[0037] In a possible implementation manner, step S140 further includes:

[0038] Step S141: Obtain the normal operating temperature threshold of the to-be-detected automotive injection molded part, and determine the starting set of abnormal regions by judging the abnormal regions in the infrared grayscale image of the surface of the standard injection molded part according to the normal operating temperature threshold.

[0039] Step S142: Select multiple seed points from the starting set of abnormal regions, and perform abnormal region growth and merging based on the multiple seed points according to the neighborhood growth rule to obtain a set of abnormal growth regions.

[0040] Step S143: Use the Sobel operator to perform edge detection calculation and edge connection and refinement on the set of abnormal growth regions to obtain a set of edge contours of the abnormal region.

[0041] Step S144: Perform edge segmentation and marking on the set of abnormal growth regions based on the set of edge contours of the abnormal region to obtain the N preliminary abnormal regions of the injection molded part.

[0042] Specifically, first, the normal operating temperature threshold of the automotive injection molded part to be detected needs to be obtained. This threshold is usually derived by collecting, analyzing, and statistically processing the temperature data of a large number of normal injection molded parts of the same type in actual usage scenarios, and then applying it to the surface infrared grayscale image of the standard injection molded part. Since there is a corresponding relationship between the grayscale value of the infrared grayscale image and the temperature, the temperature corresponding to each pixel point in the image (obtained through grayscale value conversion) is compared one by one with the normal operating temperature threshold. If the temperature of a certain pixel point exceeds the normal operating temperature threshold range, then the area where this pixel point is located is determined to be possibly abnormal. All these pixel points determined to be abnormal are aggregated, and the initial abnormal area set is thus determined.

[0043] From the initial abnormal area set, multiple seed points are selected according to a certain strategy. The selection of these seed points is not random but tends to choose those points that are representative and can effectively guide region growth, such as points with relatively significant temperature anomalies. After the seed points are selected, the neighborhood growth rule is followed. Taking each seed point as the center, check whether the temperature of its adjacent pixel points (also obtained through grayscale value conversion) is also within the abnormal range and the temperature difference from the seed point is within a certain reasonable range. If the conditions are met, the adjacent pixel point is merged into the current growth region centered on the seed point. This process is continuously repeated until there are no eligible adjacent pixel points to be merged, and thus the abnormal growth area set is obtained. Compared with the initial abnormal area set, this set more reasonably integrates the areas that may have defects.

[0044] For the obtained set of abnormally growing regions, edge detection calculations are performed using the Sobel operator. The Sobel operator includes convolution kernels in the horizontal and vertical directions. When processing each regional image in the set of abnormally growing regions, the horizontal convolution kernel performs a convolution operation on the pixel values of the image in the horizontal direction, while the vertical convolution kernel performs a convolution in the vertical direction. Through the convolution calculation, the gradient values of each pixel point in the horizontal and vertical directions are obtained, thereby reflecting the change in the gray value of the pixel point. Pixel points with larger gradient values mean that the gray change at their locations is more obvious and are more likely to be the edge pixels of the region. After calculating the gradient values of all pixel points, a set of possible edge pixel points is initially determined. However, these edge pixel points may be discontinuous and scattered, and cannot accurately outline the complete edge of the abnormal region. Therefore, an edge connection algorithm is used to process these preliminary edge pixel points. This algorithm will connect pixel points that are close in distance and have similar gradient directions based on the spatial position relationship and gradient similarity between the edge pixel points to form continuous edge line segments. At the same time, in order to make the edge more accurate, an edge thinning operation is also performed to remove some redundant edge pixels and only retain the pixel points that best represent the edge of the abnormal region. After this series of edge detection calculations, connection, and thinning operations, a clear and accurate set of edge contours of the abnormal region is obtained, providing a key basis for subsequent precise division and marking of the abnormal region.

[0045] Based on the obtained set of edge contours of the abnormal region, the set of abnormally growing regions is segmented and marked. According to the information in the set of edge contours, each independent abnormally growing region is accurately divided and a corresponding mark is added. After such processing, N preliminary abnormal regions of the injection molded parts are finally obtained.

[0046] In a possible implementation manner, step S300 further includes:

[0047] Step S310: Perform denoising preprocessing and feature point extraction on the N multi-source detection images in sequence to obtain N sets of multi-source image feature points.

[0048] Step S320: Calculate the similarity information of the N sets of multi-source image feature points, and obtain a set of matching feature points according to the similarity information.

[0049] Step S330: Based on the set of matching feature points, perform spatial transformation and registration alignment on the N multi-source detection images to obtain N multi-source registered images.

[0050] Step S340: Perform multi-scale division and fusion on the N multi-source registered images respectively to obtain N region multi-source fusion images.

[0051] Specifically, the processing is carried out sequentially for the obtained N multi-source detection images (including visible light images of injection molded parts and X-ray images of injection molded parts). Since the images will inevitably be disturbed by noise during the acquisition process, denoising preprocessing is first performed. Appropriate denoising algorithms, such as Gaussian filtering, median filtering, etc., are used to remove the noise points in the images, making the images clearer and avoiding the influence of noise on the subsequent feature point extraction. After denoising, using feature point extraction algorithms, such as SIFT (Scale-Invariant Feature Transform), SURF (Speeded-Up Robust Features), etc., representative feature points are extracted from each multi-source detection image. These feature points contain key information of the images, such as corner points, edge turning points, etc., and finally N multi-source image feature point sets are obtained.

[0052] For the obtained N multi-source image feature point sets, similarity calculation is started to obtain the matching feature point set. For each multi-source image feature point set, the feature points therein contain specific description information, such as the 128-dimensional vector descriptor of SIFT feature points. These descriptors detail the feature information such as the gradient direction and amplitude in the area around the feature points. When calculating the similarity, distance measurement methods are usually used, such as the Euclidean distance. For the descriptor vectors of two feature points, calculate the sum of the squares of the differences in their corresponding dimensions, and then take the square root. The smaller the obtained value, the higher the similarity between the two feature points. For each feature point in the N multi-source image feature point sets, the similarity is calculated one by one with all the feature points in other sets. After all the similarity calculations are completed, screening is carried out according to a pre-set similarity threshold. The feature point pairs with similarity higher than the threshold are selected, and these feature point pairs constitute the matching feature point set.

[0053] Based on the obtained matching feature point set, spatial transformation and registration alignment operations are performed on the N multi-source detection images. Using the coordinate information of the matching feature point pairs, the transformation matrix between the images is calculated, such as transformation parameters of translation, rotation, scaling, etc. Through these transformation parameters, the multi-source detection images of different modalities are subjected to corresponding spatial transformations to accurately align them in spatial positions, ensuring that the same objects or regions in the images can coincide. After such processing, N multi-source registered images are obtained. At this time, the images of different modalities already have an accurate corresponding relationship in the same coordinate system.

[0054] An image pyramid is constructed for each multi-source registered image. Through Gaussian blurring and downsampling operations, image layers with different resolutions are generated respectively to form a Gaussian pyramid. Gaussian blurring is used to smooth the image and reduce high-frequency noise, while downsampling reduces the image resolution, thus obtaining a series of images with different scales. Then, multi-scale region division is performed on different layers of each Gaussian pyramid, and the image is segmented into multiple regions with different sizes and positions, obtaining multiple sets of regions with different scales. Next, according to the pre-set fusion rules, the energy or other feature indicators of each region are calculated. These rules usually determine the weights of different regions based on the pixel values, gradient information or other statistical features of the image. Based on these weights, the corresponding regions of different modality images at the same scale are weighted and fused. For example, for the regions of visible light image and X-ray image at the same scale, the pixel values are weighted and averaged according to their weight assignments, so that the fused region can retain both the texture detail information of the visible light image and the internal structure information of the X-ray image. Finally, the fused regions with different scales are reconstructed according to the pyramid structure to obtain N multi-source fused images of regions.

[0055] In a possible implementation manner, step S340 further includes:

[0056] Step S341: Perform Gaussian blurring processing and downsampling processing on the N multi-source registered images respectively, and iteratively construct N visible light image pyramids and N X-ray image pyramids.

[0057] Step S342: Perform multi-scale region division on each layer of the N visible light image pyramids and N X-ray image pyramids to obtain N sets of visible light multi-scale regions and N sets of X-ray multi-scale regions.

[0058] Step S343: Obtain the region energy weighted fusion rule according to the requirements of the image fusion task.

[0059] Step S344: Based on the region energy weighted fusion rule, perform multi-scale fusion on the N sets of visible light multi-scale regions and N sets of X-ray multi-scale regions respectively to obtain N multi-source fused images of regions.

[0060] Specifically, the visible light images and X-ray images in the N multi-source registration images are processed separately. For each visible light image and X-ray image, first, Gaussian blur processing is performed. Utilizing the characteristics of the Gaussian function, the pixels in the image and their neighboring pixels are weighted and averaged, effectively smoothing the image, suppressing noise, and making the image details more delicate. Immediately afterwards, downsampling processing is carried out. By reducing the resolution of the image and the number of pixels, an image with a smaller size is obtained. This process of Gaussian blur and downsampling is repeated, starting from the original image, to gradually construct an image pyramid structure layer by layer. After such iterative operations, N visible light image pyramids and N X-ray image pyramids are successfully constructed, laying a foundation for analyzing image information at different scales in the subsequent steps.

[0061] On each layer of the constructed N visible light image pyramids and N X-ray image pyramids, multi-scale region division work is carried out. According to the pre-set division strategy, each layer of the image is segmented into multiple regions of different sizes and positions. For each layer of the visible light image pyramid, a series of regions of different scales are obtained after division, and these regions form N visible light multi-scale region sets; similarly, for each layer of the X-ray image pyramid, division is carried out to obtain N X-ray multi-scale region sets. These multi-scale region sets contain the local feature information of the image at different scales and are the key data for subsequent fusion processing.

[0062] Clarify the core requirements of this image fusion task, such as whether it is more focused on highlighting the surface texture details of the injection molded part or paying more attention to its internal structural features. Based on this requirement, select an appropriate energy measurement method to calculate the energy of each region in different source images (such as visible light images and X-ray images). By calculating the sum of the squares of the pixel values in the region, the sum of the gradient magnitudes, etc., to quantify the region energy, and thus obtain the energy value of each region. Then, according to the importance of different region energies in reflecting the defect information of the injection molded part, determine their weights in the fusion process. If the task is more concerned with the internal structure, the weight of the X-ray image region energy may be relatively high; on the contrary, if more attention is paid to the surface features, the weight of the visible light image region energy will be more prominent. Finally, a set of rules for weighted summation of regions according to the energy ratio is formed. According to this rule, in the subsequent fusion steps, the visible light multi-scale region set and the X-ray multi-scale region set are processed, so that the fused image can best meet the requirements of detecting the defects of the injection molded part, providing strong support for accurately identifying and locating the defects in the subsequent steps.

[0063] Based on the determined regional energy weighted fusion rule, multi-scale fusion is performed on N visible light multi-scale region sets and N X-ray multi-scale region sets. For the corresponding regions at each scale, according to the weighted fusion rule, the energies of the visible light region and the X-ray region are weighted and summed to obtain the fused region. From the bottom layer to the top layer of the pyramid, the regions at each scale are successively fused. After the fusion of all scale regions is completed, the fused regions are combined and reconstructed according to the pyramid structure, and finally N multi-source fused images of regions are obtained. These fused images fully integrate the advantageous information of the visible light image and the X-ray image at different scales, providing rich and high-quality data support for the subsequent accurate identification and positioning of the defects of the injection molded parts, and greatly improving the accuracy and reliability of defect detection.

[0064] In a possible implementation manner, step S344 further includes:

[0065] Step S3441: Define a regional energy metric index set, and calculate the regional energy of the N visible light multi-scale region sets and the N X-ray multi-scale region sets respectively according to the regional energy metric index set, to obtain N multi-scale visible light regional energy sets and N multi-scale X-ray regional energy sets.

[0066] Step S3442: Based on the regional energy weighted fusion rule, according to the regional energy ratio of the N multi-scale visible light regional energy sets and the N multi-scale X-ray regional energy sets, perform regional weighted layer-by-layer fusion on the N visible light multi-scale region sets and the N X-ray multi-scale region sets respectively, to obtain N fused image pyramids.

[0067] Step S3443: Perform inverse transformation and image reconstruction on the N fused image pyramids respectively, to obtain the N multi-source fused images of regions.

[0068] Specifically, according to the characteristics and requirements of the image fusion task, a regional energy metric set is defined. This metric set combines a variety of calculation methods that can effectively reflect the characteristics of image regions to comprehensively measure regional energy. For example, the gray-level co-occurrence matrix can be used to analyze the spatial correlation of pixel gray-level values in an image and reflect regional energy from the texture perspective; the gradient magnitude calculation can highlight the regions with drastic gray-level changes in the image, reflect the contrast information of the regions, and then measure their energy; the Laplace operator enhances image edges and provides a supplement to the measurement of regional energy. After completing the definition of the metric set, for the N visible-light multi-scale region sets and N X-ray multi-scale region sets obtained in the previous steps, the methods in the regional energy metric set are used for calculation respectively. For each region in the N visible-light multi-scale region sets, using the above methods, features such as brightness, contrast, and texture complexity within the region are quantitatively calculated to obtain the energy value of each visible-light region, and all these energy values form an N multi-scale visible-light region energy set. Similarly, for each region in the N X-ray multi-scale region sets, the same operation is performed, and the energy values are calculated using the metric set method to form an N multi-scale X-ray region energy set. These sets provide accurate data support for subsequent image fusion operations based on energy ratios, helping to more reasonably fuse the information of visible-light images and X-ray images to achieve precise detection of injection molding part defects.

[0069] After completing the energy calculation for the N visible-light multi-scale region sets and N X-ray multi-scale region sets to obtain the corresponding energy sets, based on the obtained regional energy weighted fusion rule, according to the energy ratios of each region in the N multi-scale visible-light region energy sets and N multi-scale X-ray region energy sets, the multi-scale region sets are weighted and fused layer by layer. Starting from the bottom layer of the image pyramid, for each layer of visible-light multi-scale regions and the corresponding X-ray multi-scale regions, the weights are determined according to the proportion of their respective energies in the total energy of that layer. Regions with a higher energy proportion have a greater weight during fusion, meaning they contribute more to the fusion result. According to these weights, weighted operations are performed on the visible-light regions and X-ray regions. For example, the pixel values or other features of the corresponding regions are weighted and summed according to the weights. After completing the fusion of one layer, the fusion result is used as a new layer, and the fusion of each layer is performed upward in the same way in turn. Through such layer-by-layer weighted fusion operations, finally, N fusion image pyramids are constructed. These fusion image pyramids integrate the information of visible-light images and X-ray images at different scales, laying a foundation for generating regional multi-source fusion images in subsequent further processing, helping to more comprehensively and accurately present the characteristics of injection molding parts, and thus improving the accuracy of defect detection for injection molding parts.

[0070] After completing the construction of the N fused image pyramids, the fused image pyramids are converted into the final N-region multi-source fused images. Since the images have undergone operations such as Gaussian blur and downsampling during the construction of the fused image pyramids, the image resolution is reduced and some high-frequency information is lost. Therefore, it is necessary to perform an inverse transformation on each fused image pyramid to restore the original size and resolution of the image. The inverse transformation process is opposite to the downsampling operation when constructing the pyramid. Through upsampling and corresponding filtering processes, the details of the image are gradually restored. After completing the inverse transformation, image reconstruction is performed. This process combines the fused regions at different scales according to the information of each layer of the fused image pyramid. In the order from the top layer to the bottom layer of the pyramid, the image data of each layer is integrated so that the image is complete and continuous in space. Through such inverse transformation and image reconstruction operations, the advantageous information of the visible light image and the X-ray image is fully fused, and finally N-region multi-source fused images are obtained, providing high-quality data support for the subsequent accurate identification and localization of the defects of the injection molded parts.

[0071] In a possible implementation manner, step S300 further includes:

[0072] Step S350: Obtain an automotive injection molded part defect image database, perform defect labeling and recognition training on the automotive injection molded part defect image database, and construct an injection molded part defect classifier.

[0073] Step S360: Use the injection molded part defect classifier to sequentially perform defect identification and classification on the N-region multi-source fused images to obtain an injection molded part defect feature type set.

[0074] Step S370: Perform anchor box positioning on each defect feature in the injection molded part defect feature type set respectively to obtain an injection molded part defect feature position set.

[0075] Step S380: Correlate and integrate the injection molded part defect feature type set and the injection molded part defect feature position set to determine the injection molded part region defect feature set.

[0076] Specifically, a convolutional neural network (CNN) algorithm is used to construct an injection molding part defect classifier. First, a database containing a large number of defect images of automotive injection molding parts is obtained. These images cover various types of defects, such as cracks, holes, bubbles, etc. Then, the images in the database are defect-identified, and each image is labeled with its corresponding defect type to form a labeled training dataset. Next, a convolutional neural network model is built, which usually includes multiple convolutional layers, pooling layers, and fully connected layers. The convolutional layer performs convolutional operations by sliding the convolutional kernel over the image to extract features in the image. Different convolutional kernels can capture different texture, shape, and other feature information. The pooling layer then downsamples the output of the convolutional layer, reducing the amount of data while retaining the main features. The labeled dataset of automotive injection molding part defect images is input into the CNN model for training. During the training process, the model continuously adjusts its own parameters (such as the weights and biases of the convolutional kernels), calculates the error between the predicted result and the true label through the backpropagation algorithm, and optimizes the model parameters according to the error, enabling the model to gradually learn the feature patterns of different defect types in the image. After multiple rounds of training, the model's ability to identify different defects is continuously improved, and finally a classifier that can accurately identify injection molding part defects is constructed, laying a foundation for subsequent defect identification and classification of region multi-source fusion images.

[0077] The previously obtained N region multi-source fusion images are sequentially input into the constructed injection molding part defect classifier. The classifier analyzes each fusion image and, based on the defect feature patterns it has learned, identifies the defect types present in the image, ultimately obtaining a set of injection molding part defect feature types, which covers various defect type information contained in the N fusion images.

[0078] For each defect feature in the set of injection molding part defect feature types, a target detection algorithm is used for processing. Common ones are target detection algorithms based on deep learning, such as the YOLO (You Only Look Once) series of algorithms. These algorithms search for and identify specific defect features in the input region multi-source fusion images. By extracting the features of the images, the algorithms convert the images into feature maps containing rich information, and then predict the possible positions of defects on the feature maps. For each predicted defect position, the algorithm generates a rectangular box, that is, an anchor box. The size, position, and ratio of the anchor box are generated according to pre-set rules, with the aim of enclosing the defect as accurately as possible. The algorithm continuously adjusts the parameters of the anchor box to make it match the actual defect area as much as possible. After the algorithm traverses all the region multi-source fusion images containing different defect features, a series of anchor boxes corresponding to different defect features are obtained. The position information of these anchor boxes is summarized to form a set of injection molding part defect feature positions.

[0079] Associate and integrate the injection molding part defect feature type set and the injection molding part defect feature position set. Combine the type information of the defect and its corresponding position information to form a complete data set, thereby determining the injection molding part regional defect feature set.

[0080] In a possible implementation manner, step S400 further includes:

[0081] Step S410: Analyze the operating conditions of the injection molding part to obtain the injection molding part simulation load parameters.

[0082] Step S420: Apply the injection molding part simulation load parameters to the injection molding part finite element model for solution simulation analysis to obtain the injection molding part internal defect expansion parameters.

[0083] Step S430: Evaluate the performance impact of the injection molding part internal defect expansion parameters based on the defect impact evaluation index set to generate the injection molding part internal defect detection result.

[0084] Specifically, comprehensively collect the operating condition information of the injection molding part on the vehicle, which covers multiple aspects, such as the vibration environment where the injection molding part is located during vehicle driving, including vibration frequency and amplitude; the temperature change range during vehicle operation, from low temperature at startup to high temperature after long-term driving; and the pressure borne by the injection molding part under different driving conditions, such as dynamic pressure generated during acceleration, deceleration, and turning. Then, deeply analyze these collected operating conditions, and according to relevant theories such as material mechanics and thermodynamics, convert the actual physical conditions into parameters suitable for finite element simulation. For example, convert vibration data into corresponding dynamic load parameters, temperature changes into thermal load parameters, pressure data into mechanical load parameters, etc., and finally obtain the injection molding part simulation load parameters, which can accurately simulate various physical effects borne by the injection molding part in the real use scenario, providing key input data for subsequent simulation analysis in the finite element model and ensuring that the simulation results can truly reflect the actual situation of the injection molding part.

[0085] Accurately apply the injection molded part simulation load parameters obtained in the previous step to the finite element model established based on the regional defect feature set of the injection molded part. The finite element model divides the injection molded part into numerous tiny units and simulates the overall performance by performing mechanical analysis on each unit. When applying the simulation load parameters, according to the type and characteristics of the parameters, accurately set their acting modes and positions. For example, for pressure parameters, determine their acting surfaces and directions on the injection molded part model; for temperature parameters, set the temperature distribution of each part of the model. After completing the parameter application, use professional finite element analysis software, such as ANSYS, to perform the solution simulation analysis. The software will calculate each unit in the model according to the physical properties, mechanical characteristics of the material, and the set boundary conditions. During the calculation process, simulate the mechanical response of the injection molded part under actual use conditions, and analyze the distribution of stress and strain inside the injection molded part, especially the changes in the defect area. As the simulation progresses, observe the development of defects under the action of the simulation load, such as whether cracks expand and whether holes deform. Through analysis, finally obtain the internal defect expansion parameters of the injection molded part, including key information such as the expansion speed, expansion direction, expansion length of the defect, and the resulting structural deformation amount of the injection molded part, providing an important basis for subsequent evaluation of the impact of defects on the performance of the injection molded part.

[0086] Use the pre-set defect impact evaluation index set to comprehensively evaluate the performance impact of these parameters, thereby generating the internal defect detection results of the injection molded part. The defect impact evaluation index set covers multiple key indicators for measuring the performance of the injection molded part, such as strength, stiffness, fatigue life, etc. For the internal defect expansion parameters of the injection molded part, such as the length and speed of defect expansion and the resulting structural deformation amount, etc., conduct comparative analysis with each evaluation index. If the defect expansion causes local stress concentration in the injection molded part and the stress value exceeds the allowable stress of the material, it indicates that the defect has a greater impact on the strength performance of the injection molded part; if the deformation amount of the injection molded part exceeds the allowable range due to defect expansion, it means that the stiffness performance is affected; for fatigue life, judge whether it will shorten the fatigue life of the injection molded part under normal use conditions according to the defect expansion situation and the fatigue characteristics of the material. Integrate the comprehensive evaluation results according to the established rules and logic, and finally generate the internal defect detection results of the injection molded part. This result details the state of the internal defects of the injection molded part, including the type, location, expansion situation of the defects, and the specific impact of these defects on the performance of the injection molded part.

[0087] Embodiment 2, based on the same inventive concept as the internal defect detection method for an automotive injection molded part in the foregoing embodiment, as Figure 2 shown, the present application provides an internal defect detection system for an automotive injection molded part. The system in the embodiments of the present application and the method embodiments are based on the same inventive concept. Among them, the system includes:

[0088] The preliminary injection molded part abnormal area acquisition module 10 is used to scan and image the to-be-detected automotive injection molded part using an infrared thermal imager, generate a thermal imaging map of the injection molded part surface, identify and mark the abnormal areas on the thermal imaging map of the injection molded part surface, and obtain N preliminary injection molded part abnormal areas.

[0089] The multi-source detection image acquisition module 20 is used to activate the secondary defect detection device, and collect and obtain N multi-source detection images of the N preliminary injection molded part abnormal areas through the secondary defect detection device. The N multi-source detection images include the visible light image of the injection molded part and the X-ray image of the injection molded part.

[0090] The injection molded part area defect feature set determination module 30 is used to register, align and multi-scale fuse the N multi-source detection images to obtain N area multi-source fused images, and perform defect classification and positioning on the N area multi-source fused images to determine the injection molded part area defect feature set.

[0091] The defect detection result generation module 40 is used to perform finite element simulation based on the injection molded part area defect feature set, establish a finite element model of the injection molded part, perform solution simulation analysis and defect influence evaluation on the finite element model of the injection molded part according to the use conditions of the injection molded part, and generate the internal defect detection result of the injection molded part.

[0092] Furthermore, the system is also used to implement the following functions:

[0093] Extract and obtain the distribution temperature value information of the thermal imaging map of the injection molded part surface, convert the distribution temperature value information into a grayscale image to obtain the infrared grayscale image of the injection molded part surface; according to the noise characteristic information of the thermal imaging map of the injection molded part surface, initialize the Gaussian filter, and use the Gaussian filter to filter and denoise the infrared grayscale image of the injection molded part surface to obtain the available infrared grayscale image of the injection molded part surface; based on the application requirements of image contrast, set the target pixel value threshold, map the available infrared grayscale image of the injection molded part surface to the target pixel value threshold for contrast enhancement to obtain the standard infrared grayscale image of the injection molded part surface; perform abnormal area identification and edge segmentation marking on the standard infrared grayscale image of the injection molded part surface to obtain N preliminary injection molded part abnormal areas.

[0094] Furthermore, the system is also used to implement the following functions:

[0095] Obtain the normal operating temperature threshold of the automotive injection molding part to be detected, determine the abnormal area in the standard injection molding part surface infrared grayscale image according to the normal operating temperature threshold, and determine the initial abnormal area set; select multiple seed points from the initial abnormal area set, and perform abnormal area growth and merging based on the multiple seed points according to the neighborhood growth rule to obtain the abnormal growth area set; use the Sobel operator to perform edge detection calculation and edge connection refinement on the abnormal growth area set to obtain the abnormal area edge contour set; perform edge segmentation and marking on the abnormal growth area set based on the abnormal area edge contour set to obtain the N initial injection molding part abnormal areas.

[0096] Further, the system is also used to implement the following functions:

[0097] Perform denoising preprocessing and feature point extraction on the N multi-source detection images in sequence to obtain N multi-source image feature point sets; calculate the similarity information of the N multi-source image feature point sets, and obtain the matching feature point set according to the similarity information; perform spatial transformation and registration alignment on the N multi-source detection images based on the matching feature point set to obtain N multi-source registered images; perform multi-scale division and fusion on the N multi-source registered images respectively to obtain N regional multi-source fusion images.

[0098] Further, the system is also used to implement the following functions:

[0099] Perform Gaussian blur processing and downsampling processing on the N multi-source registered images respectively, and iteratively construct N visible light image pyramids and N X-ray image pyramids; perform multi-scale region division on each layer of the N visible light image pyramids and N X-ray image pyramids to obtain N visible light multi-scale region sets and N X-ray multi-scale region sets; according to the image fusion task requirements, obtain the regional energy weighted fusion rule; perform multi-scale fusion on the N visible light multi-scale region sets and N X-ray multi-scale region sets respectively based on the regional energy weighted fusion rule to obtain N regional multi-source fusion images.

[0100] Further, the system is also used to implement the following functions:

[0101] Define a set of regional energy measurement metrics, and calculate the regional energy of the N visible light multi-scale region sets and N X-ray multi-scale region sets respectively according to the set of regional energy measurement metrics, to obtain N multi-scale visible light region energy sets and N multi-scale X-ray region energy sets; Based on the regional energy weighted fusion rule, perform regional weighted layer-by-layer fusion on the N multi-scale visible light region energy sets and N multi-scale X-ray region energy sets respectively according to the regional energy ratio, to obtain N fusion image pyramids; Perform inverse transformation and image reconstruction on the N fusion image pyramids respectively to obtain the N regional multi-source fusion images.

[0102] Further, the system is also used to implement the following functions:

[0103] Obtain an automotive injection molding part defect image database, perform defect identification and recognition training on the automotive injection molding part defect image database, and construct an injection molding part defect classifier; Use the injection molding part defect classifier to perform defect identification and classification on the N regional multi-source fusion images in sequence to obtain an injection molding part defect feature type set; Perform anchor box positioning on each defect feature in the injection molding part defect feature type set respectively to obtain an injection molding part defect feature position set; Associate and integrate the injection molding part defect feature type set and the injection molding part defect feature position set to determine the injection molding part regional defect feature set.

[0104] Further, the system is also used to implement the following functions:

[0105] Analyze the operating conditions of the injection molding part to obtain injection molding part simulation load parameters; Apply the injection molding part simulation load parameters to the injection molding part finite element model for solution simulation analysis to obtain injection molding part internal defect expansion parameters; Based on a set of defect impact evaluation metrics, perform performance impact evaluation on the injection molding part internal defect expansion parameters to generate the injection molding part internal defect detection results.

[0106] Embodiment 3. Based on the same inventive concept as the internal defect detection method of an automotive injection molding part in the foregoing embodiment, this embodiment provides a computer-readable storage medium, which can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the internal defect detection method of an automotive injection molding part in an embodiment of the present application. The processor executes various functional applications and data processing of the computer device by running the software programs, instructions, and modules stored in the memory, that is, to implement the above-mentioned internal defect detection method of an automotive injection molding part.

[0107] It should be noted that the above order of the embodiments of the present application is only for description and does not represent the superiority or inferiority of the embodiments. In addition, the above specific embodiments of this specification have been described. Further, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0108] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

[0109] This specification and the drawings are only exemplary descriptions of the present application and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. An internal defect detection method for automotive injection molded parts, characterized in that, The method includes: Scanning and imaging the injection-molded parts of the vehicle to be detected using an infrared thermal imager to generate a thermal imaging map of the surface of the injection-molded parts, identifying and marking abnormal areas on the thermal imaging map of the surface of the injection-molded parts to obtain N preliminary abnormal areas of the injection-molded parts; Activating the secondary defect detection device, and acquiring N multi-source detection images of the N preliminary abnormal areas of the injection-molded parts through the secondary defect detection device, where the N multi-source detection images include a visible light image of the injection-molded part and an X-ray image of the injection-molded part; Registering, aligning, and multi-scale fusing the N multi-source detection images to obtain N region multi-source fusion images, and classifying and locating defects in the N region multi-source fusion images to determine the region defect feature set of the injection-molded parts; Performing finite element simulation based on the region defect feature set of the injection-molded parts, establishing a finite element model of the injection-molded parts, solving, simulating, and analyzing, and evaluating the defect impact according to the usage conditions of the injection-molded parts to generate the internal defect detection result of the injection-molded parts; The obtaining of the N region multi-source fusion images includes: Performing denoising preprocessing and feature point extraction on the N multi-source detection images in sequence to obtain N multi-source image feature point sets; Calculating the similarity information of the N multi-source image feature point sets, and obtaining a set of matching feature points according to the similarity information; Performing spatial transformation and registration alignment on the N multi-source detection images based on the set of matching feature points to obtain N multi-source registered images; Performing multi-scale partitioning and fusion on the N multi-source registered images respectively to obtain N region multi-source fusion images.

2. The internal defect detection method for an automotive injection molded part according to claim 1, wherein, The obtaining of the N preliminary abnormal areas of the injection-molded parts includes: Extracting and obtaining the distributed temperature value information of the thermal imaging map of the surface of the injection-molded parts, converting the distributed temperature value information into a grayscale image to obtain an infrared grayscale image of the surface of the injection-molded parts; Initializing a Gaussian filter according to the noise characteristic information of the thermal imaging map of the surface of the injection-molded parts, and using the Gaussian filter to filter and denoise the infrared grayscale image of the surface of the injection-molded parts to obtain an available infrared grayscale image of the surface of the injection-molded parts; Setting a target pixel value threshold based on the application requirement of image contrast, mapping the available infrared grayscale image of the surface of the injection-molded parts to the target pixel value threshold for contrast enhancement to obtain a standard infrared grayscale image of the surface of the injection-molded parts; Identifying and edge-segmenting and marking abnormal areas on the standard infrared grayscale image of the surface of the injection-molded parts to obtain N preliminary abnormal areas of the injection-molded parts.

3. The internal defect detection method of an automotive injection molding part according to claim 2, characterized in that The obtaining of the N preliminary abnormal areas of the injection-molded parts includes: Obtaining the normal use temperature threshold of the injection-molded parts of the vehicle to be detected, and determining the starting abnormal area set by judging abnormal areas on the standard infrared grayscale image of the surface of the injection-molded parts according to the normal use temperature threshold; Selecting multiple seed points from the starting abnormal area set, and performing abnormal area growth and merging based on the multiple seed points according to the neighborhood growth rule to obtain an abnormal growth area set; Performing edge detection calculation and edge connection refinement on the abnormal growth area set using the Sobel operator to obtain an abnormal area edge contour set; Based on the set of abnormal region edge contours, perform edge segmentation and marking on the set of abnormal growth regions to obtain the N preliminary injection part abnormal regions.

4. The internal defect detection method for an automotive injection molded part according to claim 1, characterized in that, The obtaining of the N region multi-source fusion images includes: Perform Gaussian blur processing and downsampling processing on the N multi-source registration images respectively, and iteratively construct N visible light image pyramids and N X-ray image pyramids; Perform multi-scale region division on each layer of the N visible light image pyramids and N X-ray image pyramids to obtain N visible light multi-scale region sets and N X-ray multi-scale region sets; According to the requirements of the image fusion task, obtain the region energy weighted fusion rule; Based on the region energy weighted fusion rule, perform multi-scale fusion on the N visible light multi-scale region sets and N X-ray multi-scale region sets respectively to obtain N region multi-source fusion images.

5. The internal defect detection method of an automotive injection molding part according to claim 4, characterized in that, The obtaining of the N region multi-source fusion images includes: Define a region energy metric index set, and calculate the region energy of the N visible light multi-scale region sets and N X-ray multi-scale region sets respectively according to the region energy metric index set to obtain N multi-scale visible light region energy sets and N multi-scale X-ray region energy sets; Based on the region energy weighted fusion rule, perform region weighted layer-by-layer fusion on the N visible light multi-scale region sets and N X-ray multi-scale region sets respectively according to the region energy ratio of the N multi-scale visible light region energy sets and N multi-scale X-ray region energy sets to obtain N fusion image pyramids; Perform inverse transformation and image reconstruction on the N fusion image pyramids respectively to obtain the N region multi-source fusion images.

6. The internal defect detection method of an automotive injection molding part according to claim 5, characterized in that The determination of the injection part region defect feature set includes: Obtain an automotive injection part defect image database, perform defect identification and recognition training on the automotive injection part defect image database, and construct an injection part defect classifier; Use the injection part defect classifier to perform defect identification and classification on the N region multi-source fusion images in sequence to obtain an injection part defect feature type set; Perform anchor box positioning on each defect feature in the injection part defect feature type set respectively to obtain an injection part defect feature position set; Associate and integrate the injection part defect feature type set and the injection part defect feature position set to determine the injection part region defect feature set.

7. The internal defect detection method of an automotive injection molding part according to claim 1, characterized in that, The generation of the internal defect detection result of the injection part includes: Analyze the conditions of the injection part operating conditions to obtain injection part simulation load parameters; Apply the injection part simulation load parameters to the injection part finite element model for solution simulation analysis to obtain injection part internal defect expansion parameters; Based on the defect impact evaluation index set, perform performance impact evaluation on the injection part internal defect expansion parameters to generate the injection part internal defect detection result.

8. An internal defect detection system for automotive injection molded parts, characterized in that, The system is used to implement the internal defect detection method of an automotive injection part according to any one of claims 1-7, and the system includes: The preliminary injection molded part abnormal area acquisition module is used to scan and image the to-be-detected automotive injection molded part using an infrared thermal imager, generate a thermal imaging map of the injection molded part surface, identify and mark abnormal areas on the thermal imaging map of the injection molded part surface, and obtain N preliminary injection molded part abnormal areas; The multi-source detection image acquisition module is used to activate the secondary defect detection device, and collect and obtain N multi-source detection images of the N preliminary injection molded part abnormal areas through the secondary defect detection device. The N multi-source detection images include a visible light image of the injection molded part and an X-ray image of the injection molded part; The injection molded part area defect feature set determination module is used to register, align, and multi-scale fuse the N multi-source detection images to obtain N area multi-source fused images, and perform defect classification and localization on the N area multi-source fused images to determine the injection molded part area defect feature set; The defect detection result generation module is used to perform finite element simulation based on the injection molded part area defect feature set, establish a finite element model of the injection molded part, perform solution simulation analysis and defect impact assessment on the finite element model of the injection molded part according to the use conditions of the injection molded part, and generate an internal defect detection result of the injection molded part.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements an internal defect detection method for an automotive injection molded part as described in any one of claims 1-7.

Images