Sneaker quality automatic detection and evaluation system based on image acquisition and identification

Abstract

The invention belongs to the technical field of sneaker detection, and particularly relates to an automatic sneaker quality detection and evaluation system based on image acquisition and recognition, which comprises a multi-view image acquisition module, an intelligent image preprocessing module, a feature deep mining module, a quality defect recognition module, a comprehensive evaluation decision module and a display alarm management end, according to the invention, the appearance of the sports shoes is comprehensively presented through the multi-view image acquisition module to avoid detection blind areas, the intelligent image preprocessing module effectively de-noising and enhances and corrects images, and the feature deep mining module comprehensively and accurately extracts feature vectors in combination with deep learning and a traditional method. The quality defect identification module adopts a hybrid model to accurately and efficiently identify the type and position of a defect, and the comprehensive evaluation decision module comprehensively and objectively evaluates the quality and gives a decision suggestion according to an identification result and a preset standard.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of sports shoes detection, in particular to a sports shoe quality automatic detection and evaluation system based on image acquisition and recognition. BACKGROUND

[0002] In the field of sports shoe production and manufacturing, quality detection is a key link to ensure that products meet standards, meet consumer demand and maintain the brand image of an enterprise. Traditional sports shoe quality detection methods mainly rely on manual visual inspection for evaluation. However, manual visual detection relies on the experience and subjective judgment of the detection personnel, and different detection personnel have different understandings and grasps of the quality standards, resulting in different quality evaluations of the same pair of sports shoes in the detection process, lack of unified and objective standards, and difficulty in ensuring the accuracy and consistency of the detection results. In the face of large-scale production, high-intensity work is easy to make the detection personnel tired, further reducing the accuracy of detection and increasing the risk of missed detection and mis-detection.

[0003] With the rapid development of science and technology, image processing and machine learning technology have made significant progress. In the field of sports shoe quality detection, image-based detection methods have gradually emerged. However, conventional image-based sports shoe quality detection technology still has detection blind spots, making it difficult to quickly and correctly capture the complex appearance features of sports shoes and accurately evaluate their quality, and unable to reasonably analyze and timely warn the collection feasibility and collection control performance of the image collection process, which is not conducive to ensuring the stable and efficient performance of the sports shoe image collection process and significantly improving the image collection quality and accuracy of quality evaluation results. Therefore, a solution is proposed. SUMMARY

[0004] The purpose of the present application is to provide a sports shoe quality automatic detection and evaluation system based on image acquisition and recognition to solve the technical defects proposed in the background art.

[0005] To achieve the above purpose, the present application provides the following technical solution: a sports shoe quality automatic detection and evaluation system based on image acquisition and recognition, comprising a multi-view image acquisition module, an intelligent image preprocessing module, a feature depth mining module, a quality defect recognition module, a comprehensive evaluation decision module and a display alarm management terminal. The multi-view image acquisition module acquires images of sports shoes from multiple different angles, and the intelligent image preprocessing module pre-processes the sports shoe images transmitted by the multi-view image acquisition module. The feature depth mining module deeply mines various features of sports shoes from the pre-processed images, including appearance shape, color distribution and texture details, and sends the extracted feature vectors to the quality defect recognition module. The quality defect identification module identifies and classifies the quality defects of the sports shoes according to the extracted feature vectors by using a pre-trained deep learning model, and determines the type and position of the defects; the comprehensive evaluation decision module comprehensively evaluates the quality of the sports shoes according to the defect identification result and in combination with a pre-set quality evaluation standard, and sends an evaluation report to the display alarm management terminal.

[0006] Further, at the beginning of detection, the sports shoes to be detected are placed on a specific rotating platform, the multi-view image acquisition module controls multiple high-resolution cameras to capture images of the sports shoes from different angles of the front, side, back, top and bottom of the sports shoes, and controls the rotating speed and angle of the rotating platform to enable the cameras to capture images of the sports shoes in different postures.

[0007] Further, the pre-processing process of the intelligent image pre-processing module includes: An adaptive median filter algorithm is used to remove salt and pepper noise and random noise in the image while preserving the edge information of the image; a histogram equalization method is used to enhance the image, adjust the gray scale distribution of the image, and improve the contrast of the image; and a pre-set standard template of sports shoes is used to perform geometric correction on the image to eliminate image distortion caused by shooting angles and camera distortion.

[0008] Further, the deep learning model used by the quality defect identification module adopts a hybrid model structure combining support vector machines and deep neural networks; the support vector machine part is used to preliminarily classify the feature vectors and divide the sports shoes into two categories: normal and suspected defects; for the sports shoes with suspected defects, the feature vectors thereof are input into the deep neural network model for further fine classification to identify the specific defect type, and the position and range of the defect are determined according to the similarity of the feature vectors.

[0009] Further, the specific operation process of the comprehensive evaluation decision module includes: The defect identification result transmitted by the quality defect identification module is received, and the severity of each defect is scored according to a pre-set quality evaluation standard; the quality evaluation standard considers the type, position and size of the defect, and different types of defects correspond to different scoring weights; The total defect score of the sports shoes is calculated according to the scores of the defects; the quality of the sports shoes is divided into different grades, including excellent, qualified and unqualified, according to the total defect score, and a detailed evaluation report is generated, including the basic information of the sports shoes, the defect type and position, the quality grade and decision suggestions.

[0010] Further, the display alarm management terminal communication connection feasibility decision module can monitor and analyze the image acquisition process, generate an acquisition influence signal or an acquisition feasible signal through analysis, and send the acquisition influence signal or the acquisition feasible signal to the display alarm management terminal. The display alarm management terminal issues a corresponding early warning when receiving the acquisition influence signal.

[0011] Further, the specific analysis process of the acquisition detection control module is as follows: The light intensity in the image acquisition environment is obtained, the light intensity is differentially calculated with the set standard brightness value and the absolute value is taken to obtain the light influence value. The dust particle concentration in the image acquisition environment is marked as the visible influence value, and the vibration amplitude of the rotating platform carrying the sports shoes is marked as the load vibration influence value. The light influence value, the visible influence value, and the load vibration influence value are respectively compared with the preset light influence threshold value, the preset visible influence threshold value, and the preset load vibration influence threshold value. If the light influence value, the visible influence value, or the load vibration influence value exceeds the corresponding preset threshold value, an acquisition influence signal is generated. If the light influence value, the visible influence value, and the load vibration influence value do not exceed the corresponding preset threshold value, the feasibility decision value is calculated by weighted summation of the light influence value, the visible influence value, and the load vibration influence value. The feasibility decision value is compared with the preset feasibility decision threshold value. If the feasibility decision value exceeds the preset feasibility decision threshold value, an acquisition influence signal is generated. If the feasibility decision value does not exceed the preset feasibility decision threshold value, an acquisition feasible signal is generated.

[0012] Further, the feasibility decision module is communicatively connected to the acquisition control evaluation module. The feasibility decision module sends the acquisition influence signal or the acquisition feasible signal to the acquisition control evaluation module. The acquisition control evaluation module analyzes the image acquisition control performance of the detection period, generates an acquisition control qualified signal or an acquisition control abnormal signal through analysis, and sends the acquisition control qualified signal or the acquisition control abnormal signal to the display alarm management terminal. The display alarm management terminal issues a corresponding early warning when receiving the acquisition control abnormal signal.

[0013] Further, the specific analysis process of the acquisition control evaluation module includes: The number of times of generation of the acquisition influence signal in the detection period is obtained and is marked as the acquisition influence feature value. The acquisition influence feature value is compared with the preset acquisition influence feature threshold value. If the acquisition influence feature value exceeds the preset acquisition influence feature threshold value, an acquisition control abnormal signal is generated. If the collected influence characteristic value does not exceed the preset collection influence characteristic threshold, the coincidence degree of the actual posture compared with the corresponding standard posture when the sports shoes switch to the corresponding posture is marked as a posture coincidence coefficient, and the posture coincidence coefficient is compared with a preset posture coincidence coefficient threshold in value. If the posture coincidence coefficient does not exceed the preset posture coincidence coefficient threshold, an abnormal posture symbol PX-1 is assigned. The number of times of assigning the abnormal posture symbol PX-1 in the detection period is obtained, and a ratio calculation is performed on the number of times of assigning the abnormal posture symbol PX-1 and the number of posture switches in the detection period to obtain a posture abnormality characteristic value. The posture abnormality characteristic value is compared with a preset posture abnormality characteristic threshold in value. If the posture abnormality characteristic value exceeds the preset posture abnormality characteristic threshold, a collection control abnormal signal is generated. If the posture abnormality characteristic value does not exceed the preset posture abnormality characteristic threshold, a rotation control evaluation value is obtained through analysis. The rotation control evaluation value is compared with a preset rotation control evaluation threshold in value. If the rotation control evaluation value exceeds the preset rotation control evaluation threshold, a collection control abnormal signal is generated. If the rotation control evaluation value does not exceed the preset rotation control evaluation threshold, a collection control qualified signal is generated.

[0014] Further, the analysis method of the rotation control evaluation value is as follows: The motion of the rotating platform during the posture switching process of the sports shoes is monitored in real time, and the real-time motion speed of the rotating platform is collected. The motion speed curve of the rotating platform during the posture switching process is obtained accordingly. A plurality of detection points are set on the motion speed curve. The speed difference value between adjacent two detection points is marked as a speed wave detection value, and the speed wave detection value is compared with a preset speed wave detection threshold in value. If the speed wave detection value exceeds the preset speed wave detection threshold, the corresponding speed wave detection value is marked as a speed wave abnormal value. The number of speed wave abnormal values in the corresponding posture switching process is obtained, and a ratio calculation is performed on the number of speed wave abnormal values and the number of speed wave detection values to obtain a speed wave abnormality value. The proportion of the time length during which the real-time motion speed is not within the corresponding preset standard speed range in the corresponding posture switching process is marked as a speed deviation time occupancy value. The actual motion trajectory in the corresponding posture switching process is obtained, and the actual motion trajectory is compared with the corresponding standard motion trajectory. The number of times of motion trajectory deviation is obtained and marked as a trajectory deviation detection value. The proportion of the non-coincidence trajectory of the actual motion trajectory compared with the corresponding standard motion trajectory is marked as a trajectory non-coincidence value. The speed wave abnormality value, the speed deviation time occupancy value, the trajectory deviation detection value, and the trajectory non-coincidence value are weighted and summed to obtain a switching monitoring coefficient. The switching monitoring coefficient is compared with a preset switching monitoring coefficient threshold in value. If the switching monitoring coefficient exceeds the preset switching monitoring coefficient threshold, the corresponding posture switching process is marked as a non-optimal switching process. The number of non-optimal switching processes in the detection period is obtained, and a ratio calculation is performed on the number of non-optimal switching processes and the number of posture switches in the detection period to obtain a rotation control evaluation value.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In this invention, by fully presenting the appearance of the sports shoe to avoid blind spots in detection, the intelligent image preprocessing module effectively denoises, enhances and corrects the image, and combines deep learning with traditional methods to fully and accurately extract feature vectors and adopt a hybrid model to accurately and efficiently identify defect types and locations. Based on the identification results and preset standards, the quality is comprehensively and objectively evaluated, thereby realizing the automation, intelligence and efficiency of sports shoe quality inspection. 2. In this invention, the feasibility decision-making module monitors and analyzes the image acquisition process and performs corresponding control operations in a timely manner to avoid the adverse effects of environmental factors on the quality of the acquired images. This significantly reduces the difficulty of supervising the acquisition of sports shoe images and further improves the accuracy of the sports shoe quality inspection and evaluation results. In addition, the acquisition control evaluation module analyzes the image acquisition control performance during the detection period and strengthens the monitoring and control of the image acquisition process when an abnormal acquisition control signal is generated, ensuring the stable and efficient operation of the sports shoe image acquisition process and improving the image acquisition quality. Attached Figure Description

[0016] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a system block diagram of Embodiment 1 of the present invention; Figure 2 This is a system block diagram of Embodiments 2 and 3 of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1: As Figure 1 As shown, the automated quality inspection and evaluation system for sports shoes based on image acquisition and recognition proposed in this invention includes a multi-view image acquisition module, an intelligent image preprocessing module, a feature deep mining module, a quality defect identification module, a comprehensive evaluation and decision module, and a display alarm management terminal. The multi-view image acquisition module captures comprehensive and clear images of the sports shoe from multiple different angles, ensuring complete information on all parts of the sports shoe. Multi-view acquisition can fully display the appearance features of the sports shoe, avoid detection blind spots caused by a single viewpoint, improve the accuracy and comprehensiveness of detection, and provide a reliable data foundation for subsequent processing. It should be noted that at the beginning of detection, the sports shoes to be detected are placed on a specific rotating platform, the multi-view image acquisition module controls multiple high-resolution cameras to take pictures from different angles such as the front, side, back, top and bottom of the sports shoes; at the same time, the rotating speed and angle of the rotating platform are controlled to enable the cameras to capture images of the sports shoes in different postures.

[0019] The intelligent image preprocessing module pre-processes the sports shoe images transmitted by the multi-view image acquisition module, including denoising, enhancement, correction and other operations, which can effectively remove noise and interference in the images, enhance the contrast and clarity of the images, correct the geometric deformation of the images, provide high-quality images for subsequent feature extraction, and improve the accuracy and reliability of feature extraction; specifically, the pre-processing process of the intelligent image preprocessing module is as follows: First, the adaptive median filtering algorithm is used to denoise the image, remove salt and pepper noise and random noise in the image, and retain the edge information of the image; then, the histogram equalization method is used to enhance the image, adjust the gray distribution of the image, improve the contrast of the image, and make the details of the sports shoes more clear and visible; then, according to the pre-set standard template of the sports shoes, the image is geometrically corrected to eliminate image deformation caused by shooting angle and camera distortion, which is conducive to ensuring the shape and position accuracy of the sports shoes in the image.

[0020] The feature depth mining module deeply mines various features of the sports shoes from the pre-processed images, including appearance shape, color distribution and texture details, and sends the extracted feature vectors to the quality defect recognition module, which combines deep learning and traditional image processing methods to comprehensively and accurately extract various features of the sports shoes, converts image information into digital feature vectors that can be processed by computers, and provides strong data support for quality defect recognition; Specifically, the feature depth mining module uses a convolutional neural network (CNN) based on deep learning to extract features from the image; the CNN model automatically learns high-level features in the image through multiple convolution layers, pooling layers and fully connected layers; in the convolution layer, different convolution kernels are used to perform convolution operations on the image to extract local features of the image such as edges and corners; The pooling layer reduces the sampling of the feature map output by the convolution layer to reduce the feature dimension and improve the computational efficiency and generalization ability of the model; the fully connected layer integrates the feature vectors output by the pooling layer to generate the final digital feature vector. At the same time, in order to more comprehensively describe the features of the sports shoes, the module will also combine traditional image processing methods such as color histogram statistics and texture feature extraction algorithms to extract color distribution and texture detail features of the sports shoes, and fuse these features with the features extracted by CNN to form a more rich feature vector.

[0021] The quality defect identification module uses a pre-trained deep learning model to identify and classify quality defects in sports shoes based on the extracted feature vectors, and determines the type and location of the defects. By adopting a hybrid model structure, it combines the fast classification capability of SVM with the powerful feature learning capability of DNN, which can accurately and efficiently identify quality defects in sports shoes and determine the type and location of the defects, providing a basis for subsequent comprehensive evaluation. Specifically, the quality defect identification module uses a hybrid model structure that combines support vector machines (SVM) and deep neural networks (DNN). The SVM part is used to perform preliminary classification of feature vectors, dividing athletic shoes into normal and suspected defect categories. For athletic shoes with suspected defects, their feature vectors are then input into the DNN model for further classification to identify specific defect types, such as glue separation, broken threads, and color difference. During the identification process, the model matches and judges based on the similarity of feature vectors to determine the location and extent of the defect.

[0022] The comprehensive evaluation and decision-making module, based on defect identification results and pre-set quality assessment standards, conducts a comprehensive quality assessment of the athletic shoes and sends the assessment report to the display and alarm management terminal. It can provide a comprehensive and objective quality assessment of the athletic shoes based on defect identification results and preset standards, and offer clear decision-making suggestions, providing strong support for the quality control and decision-making of manufacturing enterprises. The specific operation process of the comprehensive evaluation and decision-making module is as follows: The system receives the defect identification results transmitted by the quality defect identification module and scores the severity of each defect according to the pre-set quality assessment standards. The quality assessment standards take into account factors such as the type, location, and size of the defect. Different types of defects correspond to different scoring weights. For example, defects that affect the use of sports shoes, such as glue separation and broken threads, score higher, while minor appearance defects such as slight color difference score lower. Based on the scores of each defect, the total defect score of the athletic shoe is calculated. Then, the quality of the athletic shoe is divided into different levels according to the total defect score, such as excellent, qualified, and unqualified. Finally, a detailed evaluation report is generated, including the basic information of the athletic shoe, the type and location of defects, the quality level, and decision recommendations (such as whether it can be sold after leaving the factory, whether it needs to be repaired, etc.).

[0023] Example 2: Figure 2 As shown, the difference between this embodiment and Embodiment 1 is that the display alarm management terminal is connected to the feasibility decision module. The feasibility decision module monitors and analyzes the image acquisition process, generates acquisition impact signals or acquisition feasibility signals through analysis, and sends the acquisition impact signals or acquisition feasibility signals to the display alarm management terminal. When the alarm management terminal receives a signal affecting the acquisition, it issues a corresponding warning and performs timely adjustment operations to avoid adverse effects of environmental factors on the quality of the acquired images. This significantly reduces the difficulty of monitoring and acquiring images of athletic shoes and further improves the accuracy of athletic shoe quality inspection and evaluation results. The specific analysis process of the acquisition and detection control module is as follows: The illumination level in the image acquisition environment is obtained. The difference between the illumination level and the set standard brightness value is calculated and the absolute value is taken to obtain the illumination influence value. The dust particle concentration in the image acquisition environment is marked as the visible influence value, and the vibration amplitude of the rotating platform carrying the sports shoes is marked as the vibration influence value. The illumination influence value, visible influence value, and vibration influence value are compared with the preset illumination influence threshold, preset visible influence threshold, and preset vibration influence threshold, respectively. If the illumination influence value, visible influence value, or vibration influence value exceeds the corresponding preset threshold, it indicates that the current situation is not conducive to ensuring the image acquisition quality, and an acquisition influence signal is generated. If the light impact value, visible impact value, and vibration impact value do not exceed the corresponding preset threshold, the feasibility decision value is obtained by weighted summation of the light impact value, visible impact value, and vibration impact value. That is, the light impact value, visible impact value, and vibration impact value are each assigned a corresponding preset weight coefficient, and the light impact value, visible impact value, and vibration impact value are each multiplied by the corresponding preset weight coefficient. The sum of the three sets of product results is marked as the feasibility decision value. It should be noted that the larger the feasibility decision value, the more unfavorable it is to ensure image acquisition quality. The feasibility decision value is compared with the preset feasibility decision threshold. If the feasibility decision value exceeds the preset feasibility decision threshold, it indicates that it is unfavorable to ensure image acquisition quality, and an acquisition impact signal is generated. If the feasibility decision value does not exceed the preset feasibility decision threshold, it indicates that it is favorable to ensure image acquisition quality, and an acquisition feasibility signal is generated.

[0024] Example 3: Figure 2 As shown, the difference between this embodiment and Embodiment 1 and Embodiment 2 is that the feasibility decision module is connected to the acquisition control and evaluation module. The feasibility decision module sends the acquisition impact signal or acquisition feasibility signal to the acquisition control and evaluation module. The acquisition control and evaluation module analyzes the image acquisition control performance during the detection period and generates an acquisition control qualified signal or acquisition control abnormal signal through analysis. Furthermore, it sends either a qualified or abnormal acquisition and control signal to the display alarm management terminal. Upon receiving an abnormal acquisition and control signal, the display alarm management terminal issues a corresponding warning to remind management personnel to investigate and analyze the cause and strengthen the monitoring and control of the image acquisition process. This ensures the stable and efficient acquisition of sports shoe images and improves image acquisition quality, demonstrating a high level of intelligence. The specific analysis process of the acquisition and control evaluation module is as follows: The generation number of the acquisition influence signal in the detection period is obtained and marked as an acquisition influence characteristic value. The acquisition influence characteristic value is compared with a preset acquisition influence characteristic threshold value. If the acquisition influence characteristic value exceeds the preset acquisition influence characteristic threshold value, it indicates that the control performance for the environmental influence factors in the sneaker image acquisition process in the detection period is poor, and an acquisition control abnormal signal is generated. If the acquisition influence characteristic value does not exceed the preset acquisition influence characteristic threshold value, it indicates that the control performance for the environmental influence factors in the sneaker image acquisition process in the detection period is good. When the sneaker switches to the corresponding posture, the coincidence degree of the actual posture compared with the corresponding standard posture is marked as a posture coincidence coefficient. The posture coincidence coefficient is compared with a preset posture coincidence coefficient threshold value. If the posture coincidence coefficient does not exceed the preset posture coincidence coefficient threshold value, it indicates that the final presented posture effect after switching is poor, and a posture abnormal symbol PX-1 is given. The number of times of giving the posture abnormal symbol PX-1 in the detection period is obtained, and a ratio calculation is performed on the number of times of posture switching in the detection period to obtain a posture abnormal characteristic value. The posture abnormal characteristic value is compared with a preset posture abnormal characteristic threshold value. If the posture abnormal characteristic value exceeds the preset posture abnormal characteristic threshold value, it indicates that the posture switching effect in the sneaker image acquisition process in the detection period is poor, and an acquisition control abnormal signal is generated. If the posture abnormal characteristic value does not exceed the preset posture abnormal characteristic threshold value, it indicates that the posture switching effect in the sneaker image acquisition process in the detection period is good. The motion of the rotating platform is monitored in real time during the posture switching process of the sneaker. The real-time motion speed of the rotating platform is collected. The motion speed curve of the rotating platform in the posture switching process is obtained accordingly. A plurality of detection points are set on the motion speed curve. The speed difference value of adjacent two detection points is marked as a speed wave detection value. The speed wave detection value is compared with a preset speed wave detection threshold value. If the speed wave detection value exceeds the preset speed wave detection threshold value, it indicates that the speed fluctuation in the corresponding time length is too large, and the corresponding speed wave detection value is marked as a speed wave abnormal value. The number of speed wave abnormal values in the corresponding posture switching process is obtained, and a ratio calculation is performed on the number of speed wave detection values to obtain a speed wave abnormal value. The proportion of time length in which the real-time motion speed is not in the corresponding preset standard speed range in the corresponding posture switching process is marked as a speed deviation time proportion value. The actual motion trajectory in the corresponding posture switching process is obtained. The actual motion trajectory is compared with the corresponding standard motion trajectory. The number of times of motion trajectory deviation is obtained and marked as a trajectory deviation value. The proportion of non-coincidence trajectory of the actual motion trajectory compared with the corresponding standard motion trajectory (i.e. the ratio of the length of the non-coincidence trajectory to the total length of the standard motion trajectory) is marked as a trajectory non-coincidence value. The switching monitoring coefficient is calculated by weighted summation of the speed wave anomaly value, the speed time offset value, the trajectory deviation value and the trajectory non-conformity value, that is, the speed wave anomaly value, the speed time offset value, the trajectory deviation value and the trajectory non-conformity value are respectively assigned with corresponding preset weight coefficients, and the speed wave anomaly value, the speed time offset value, the trajectory deviation value and the trajectory non-conformity value are respectively multiplied by the corresponding preset weight coefficients, and the sum of the four groups of product results is marked as the switching monitoring coefficient; It should be noted that the greater the value of the switching monitoring coefficient, the worse the comprehensive motion control performance of the corresponding attitude switching process; the switching monitoring coefficient is compared with the preset switching monitoring coefficient threshold value, if the switching monitoring coefficient exceeds the preset switching monitoring coefficient threshold value, it indicates that the comprehensive motion control performance of the corresponding attitude switching process is poor, and then the corresponding attitude switching process is marked as a switching non-optimal process; The number of switching non-optimal processes in the detection period is obtained, and a ratio calculation is performed between the number of attitude switching times in the detection period to obtain a rotation control evaluation value; the rotation control evaluation value is compared with the preset rotation control evaluation threshold value, if the rotation control evaluation value exceeds the preset rotation control evaluation threshold value, it indicates that the motion control performance of the detection period for attitude switching is poor, and then a collection control abnormal signal is generated; if the rotation control evaluation value does not exceed the preset rotation control evaluation threshold value, it indicates that the collection control performance for the motion shoe image collection in the detection period is good in general, and then a collection control qualified signal is generated.

[0025] In summary, the multi-view image collection module presents the appearance of the sports shoes comprehensively to avoid detection blind spots, the intelligent image preprocessing module effectively denoises, enhances and corrects the image, the feature depth mining module comprehensively and accurately extracts the feature vector by combining deep learning and traditional methods, the quality defect recognition module adopts a hybrid model to accurately and efficiently recognize the defect type and position, and the comprehensive evaluation decision module objectively evaluates the quality and gives decision suggestions according to the recognition result and the preset standard, thereby realizing the automation, intelligence and high efficiency of sports shoe quality detection, having the advantages of high detection efficiency, high accuracy and strong objectivity compared with traditional manual detection, effectively avoiding subjective factors and missed detection and misdiagnosis problems of manual detection, and providing an advanced and reliable quality detection scheme for sports shoe production enterprises.

[0026] In the technical scheme of the present application, the threshold value or the preset value, the preset range and the like are set for result comparison and analysis to determine whether it is good or bad, and the size of the value is determined according to the large model analysis of sample data and the combination of artificial experience to set the input storage, and appropriate adjustment can be made according to the seasonal or rational influence conditions; and the preset weight coefficient and the influence factor are set according to the influence of each parameter on the result to allocate specific values to finally reflect the influence of the result, which is also set by the large model analysis of sample data and the combination of artificial experience to set the input storage, and appropriate adjustment can be made according to the seasonal or rational influence conditions.

[0027] The preferred embodiments of the application disclosed above are only used to illustrate the present application, the preferred embodiments do not describe all the details, and do not limit the application to the specific embodiments. Obviously, many modifications and variations can be made according to the content of the specification. The specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application. The application is limited only by the claims and their full scope and equivalents.

Claims

1. An automated quality inspection and evaluation system for athletic shoes based on image acquisition and recognition, characterized in that, It includes a multi-view image acquisition module, an intelligent image preprocessing module, a feature deep mining module, a quality defect identification module, a comprehensive evaluation and decision-making module, and a display alarm management terminal; The multi-view image acquisition module acquires images of the sports shoe from multiple different angles. The intelligent image preprocessing module preprocesses the sports shoe images transmitted from the multi-view image acquisition module. The feature depth mining module deeply mines various features of the sports shoe from the preprocessed images and sends the extracted feature vectors to the quality defect identification module. The quality defect identification module uses a pre-trained deep learning model to identify and classify quality defects in athletic shoes based on the extracted feature vectors, determining the type and location of the defects. The comprehensive evaluation and decision module uses the defect identification results and pre-set quality evaluation standards to comprehensively evaluate the quality of athletic shoes and sends the evaluation report to the display alarm management terminal.

2. The automated quality inspection and evaluation system for sports shoes based on image acquisition and recognition according to claim 1, characterized in that, At the start of the test, the sports shoes to be tested are placed on a specific rotating platform. The multi-view image acquisition module controls multiple high-resolution cameras to take pictures from multiple different angles of the sports shoes. By controlling the rotation speed and angle of the rotating platform, the cameras can capture images of the sports shoes in different postures.

3. The automated quality inspection and evaluation system for sports shoes based on image acquisition and recognition according to claim 1, characterized in that, The preprocessing process of the intelligent image preprocessing module includes: An adaptive median filtering algorithm is used to denoise the image; histogram equalization is used to enhance the image; and geometric correction is performed on the image based on a pre-set sports shoe standard template.

4. The automated quality inspection and evaluation system for sports shoes based on image acquisition and recognition according to claim 1, characterized in that, The deep learning model used in the quality defect identification module adopts a hybrid model structure that combines support vector machines and deep neural networks.

5. The automated quality inspection and evaluation system for athletic shoes based on image acquisition and recognition according to claim 1, characterized in that, The specific operation process of the comprehensive evaluation and decision-making module includes: receiving the defect identification results transmitted by the quality defect identification module; scoring the severity of each defect according to the pre-set quality evaluation standards; calculating the total defect score of the sports shoe based on the scores of each defect; classifying the quality of the sports shoe into different levels based on the total defect score; and generating a detailed evaluation report.

6. The automated quality inspection and evaluation system for sports shoes based on image acquisition and recognition according to claim 1, characterized in that, The display alarm management terminal communicates with the feasibility decision module, which monitors and analyzes the image acquisition process. Through analysis, it generates acquisition impact signals or acquisition feasibility signals and sends them to the display alarm management terminal.

7. The automated quality inspection and evaluation system for sports shoes based on image acquisition and recognition according to claim 6, characterized in that, The specific analysis process of the data acquisition and detection control module is as follows: If the illumination impact value, visible impact value, or vibration impact value exceeds the corresponding preset threshold, a data acquisition impact signal is generated. If none of the illumination impact value, visible impact value, or vibration impact value exceeds the corresponding preset threshold, a feasibility decision value is calculated by weighted summation of the illumination impact value, visible impact value, and vibration impact value. If the feasibility decision value exceeds the preset feasibility decision threshold, a data acquisition impact signal is generated. If the feasibility decision value does not exceed the preset feasibility decision threshold, a data acquisition feasible signal is generated.

8. The automated quality inspection and evaluation system for sports shoes based on image acquisition and recognition according to claim 6, characterized in that, The feasibility decision module communicates with the data acquisition and control evaluation module. The data acquisition and control evaluation module analyzes the image acquisition and control performance during the detection period and sends the qualified acquisition and control signal or the abnormal acquisition and control signal to the display alarm management terminal.

9. The automated quality inspection and evaluation system for sports shoes based on image acquisition and recognition according to claim 8, characterized in that, The specific analysis process of the data collection, control, and assessment module includes: The number of times the acquisition impact signal is generated during the detection period is obtained and marked as the acquisition impact feature value. If the acquisition impact feature value exceeds the preset acquisition impact feature threshold, an acquisition control abnormal signal is generated. If the acquired impact feature value does not exceed the preset acquisition impact feature threshold, the number of times the attitude anomaly symbol PX-1 is assigned during the detection period is obtained and the ratio of this number to the number of attitude switching times during the detection period is calculated to obtain the attitude anomaly feature value. If the attitude anomaly feature value exceeds the preset attitude anomaly feature threshold, an acquisition control anomaly signal is generated. If the attitude anomaly feature value does not exceed the preset attitude anomaly feature threshold, the rotation control evaluation value is obtained through analysis. If the rotation control evaluation value exceeds the preset rotation control evaluation threshold, an acquisition control anomaly signal is generated. If the rotation control evaluation value does not exceed the preset rotation control evaluation threshold, an acquisition control qualified signal is generated.

10. The automated quality inspection and evaluation system for sports shoes based on image acquisition and recognition according to claim 9, characterized in that, The specific methods for analyzing and obtaining rotation control evaluation values ​​are as follows: The switching monitoring coefficient is calculated by weighted summation of the velocity wave out-of-range value, velocity deviation time value, trajectory deviation measurement value, and trajectory non-combination value. If the switching monitoring coefficient exceeds the preset switching monitoring coefficient threshold, the corresponding attitude switching process is marked as a non-optimal switching process. The number of non-optimal switching processes during the detection period is obtained and the ratio is calculated with the number of attitude switching times during the detection period to obtain the rotation control evaluation value.

Images