Intelligent scanning platform and sorting optimization method for image self-correction

Abstract

The application provides an image self-correcting intelligent scanning platform and a sorting optimization method, and relates to the technical field of image processing.The platform comprises a multi-angle image acquisition module, a barcode repair and identification module, an image evaluation module and an image classification module.The scanning device is used to obtain image information of an object to be identified, including a barcode image and an object sorting module.The barcode repair and identification module is internally provided with an AI repair algorithm for a blurred or damaged barcode, which can repair the blurred or damaged barcode image obtained by scanning, solve the identification problem of a tape-covered or dirty barcode, improve the scanning code recognition rate, and the image classification module is internally provided with an AI classification model for an object image, which can automatically mark the type of goods during the scanning process, replace manual label pasting and classification, reduce the labor cost of a sorting center, realize efficient and accurate logistics sorting optimization, and solve the problem of sorting efficiency bottleneck caused by barcode failure in logistics storage.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more particularly to an intelligent scanning platform and sorting optimization method for image self-correction. Background Technology

[0002] In the operation of logistics warehouses, barcode scanning and sorting of goods are crucial links. However, traditional barcode scanning equipment and sorting methods face many challenges. On the one hand, barcodes are easily affected by factors such as tape covering and soiling, leading to a decrease in barcode recognition rate and affecting the efficiency of goods turnover. On the other hand, manual labeling and sorting is not only inefficient but also has high labor costs. With the rapid development of the logistics industry, the volume of goods handled is constantly increasing, and the requirements for sorting efficiency and accuracy are also getting higher and higher. Existing technologies are difficult to meet this demand: for example, multispectral scanning is used to enhance barcode recognition, but it cannot reconstruct the structure of damaged barcodes; barcode repair based on template matching has poor generalization ability and cannot adapt to complex soiled scenarios; another method is to replace barcodes with RFID, but the cost of tags increases by 200%, making it difficult to apply on a large scale.

[0003] Therefore, there is an urgent need for an intelligent scanning and sorting optimization solution that can improve barcode recognition rate and reduce labor costs. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides an intelligent scanning platform and sorting optimization method with self-correcting images. The barcode repair module incorporates an AI repair algorithm for blurry and damaged barcodes, enabling the repair of scanned blurry or damaged barcode images. This solves the identification problems of barcodes covered in tape or soiled with dirt, improving the barcode recognition rate. The image classification module incorporates an AI classification model for item images, automatically marking the type of goods during scanning, replacing manual labeling and reducing labor costs at sorting centers.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect of the present invention, an image self-correcting intelligent scanning platform is provided, comprising:

[0007] A multi-angle image acquisition module is configured above and to the side of the sorting line transport device to dynamically acquire image information of the items to be identified, including barcode images and images of the item's appearance.

[0008] A barcode repair and recognition module is connected to the multi-angle image acquisition module. The barcode repair module has a built-in barcode AI repair algorithm, which is used to repair blurry or damaged barcode images and identify items based on the repaired barcode.

[0009] An image classification module is connected to the multi-angle image acquisition module. The image classification module has a built-in AI classification model for item images, which is used to automatically mark the type of goods based on the barcode image and the appearance image of the item during the scanning process.

[0010] An image evaluation module is connected to the barcode repair and recognition module and the image classification module. The image evaluation module is used to identify the item type by combining the barcode recognition results and AI classification labels.

[0011] The item sorting module, connected to the image evaluation module, is used to generate sorting path instructions based on the item type and coordinate with the sorting robotic arm to execute them.

[0012] Preferably, the multi-angle image acquisition module includes acquisition devices configured above and to the side of the sorting line transport device. The acquisition devices are high-definition industrial cameras. Subsequently, the image information acquired by the high-definition industrial camera is subjected to basic preprocessing. The preprocessing steps include Gaussian smoothing for noise reduction and histogram equalization to enhance contrast.

[0013] Preferably, the AI ​​repair algorithm includes:

[0014] The identification unit is used to identify blurred or damaged areas in a barcode image and determine their location and extent.

[0015] The image restoration model is trained using a large number of blurry and damaged barcode images and their corresponding standard clear barcode images. Then, the trained deep learning model is used to restore the blurry or damaged barcode images, generating restored clear barcode images, and then the items are identified based on the restored clear barcode images.

[0016] More preferably, the restoration results are evaluated by introducing an image restoration accuracy index, the calculation formula of which is as follows:

[0017] ,

[0018] in, Indicates the accuracy index of image restoration. Indicates the deviation coefficient. This represents the ground-corrected image with position coordinates (i, j) on the image. This represents the output image with position coordinates (i, j) on the image. Indicates Centered on, along its main migration direction The larger the deviation of the extended linear neighborhood window, the greater the fluctuation of the shadow of that point within its directional neighborhood, which does not conform to the principle of diffusion path stationarity. N represents the number of samples, an image restoration accuracy metric. The closer the value is to 0, the higher the repair accuracy.

[0019] Preferably, the item image AI classification model includes:

[0020] The feature extraction unit is used to extract key features from product appearance images, including shape, color, and texture;

[0021] The classification training unit, including a deep learning classification model, is trained using a large number of labeled item images and can accurately identify different types of goods.

[0022] The classification execution unit is used to classify and identify newly scanned images of the appearance of items based on a trained deep learning classification model, automatically label the type of goods, generate classification result information, and store it in association with the corresponding barcode information.

[0023] Preferably, the image evaluation module identifies the type of item through the following steps:

[0024] When the confidence level for barcode recognition is greater than 90%, barcode information should be used preferentially.

[0025] When the confidence level of barcode recognition is less than or equal to 90%, a weighted decision is made by combining the item classification results with the barcode information;

[0026] Weighted decision is calculated using the following formula:

[0027] ,

[0028] in, Indicates weighted decision-making. These are weighting coefficients. For the barcode recognition results of the item, The image recognition results for the item.

[0029] Preferably, the item sorting module includes a sorting robotic arm, which sorts items to corresponding target areas according to item type.

[0030] In a second aspect of the invention, a sorting optimization method is also provided, comprising the following steps:

[0031] S1: Dynamically acquire multi-view images of items in transit using industrial high-definition cameras configured above and to the side of the sorting line transport device;

[0032] S2: Perform an integrity assessment on the barcode area. If there is any blurring or damage, repair and recognize the barcode using a barcode AI repair algorithm.

[0033] S3: Parallel execution of barcode recognition and item image classification dual-channel analysis, wherein the item image classification is achieved through an image AI classification model;

[0034] S4: Combines barcode recognition results and AI classification labels to identify the item type;

[0035] S5: Maps the recognition results to the sorting instruction set, driving the sorting robotic arm to automatically sort items.

[0036] The beneficial effects of this invention are as follows:

[0037] 1. Improve barcode recognition rate: Through the built-in AI repair algorithm for blurry and damaged barcodes, it can effectively solve the problem of recognizing barcodes covered by tape and soiled in logistics warehouses, significantly improve the barcode recognition rate, ensure the accurate acquisition of cargo information, reduce cargo processing delays caused by barcode problems, and improve the overall operational efficiency of logistics warehouses.

[0038] 2. Reduce labor costs: By utilizing the built-in AI classification model for item images, the system automatically marks the type of goods while scanning them, replacing the traditional manual labeling and classification work. This greatly reduces the manpower input in the sorting center, lowers labor costs, and improves the accuracy and consistency of classification, avoiding errors that may occur during manual classification.

[0039] 3. Optimize the sorting process: The repaired barcode information and automatically marked goods type information are promptly sent to the sorting control system. The sorting control system can then use this information to plan sorting paths and allocate tasks accurately, achieving efficient and accurate logistics sorting optimization, improving sorting efficiency, reducing the time goods spend in the warehouse, accelerating the turnover of goods, and enhancing customer satisfaction.

[0040] 4. Enhanced Intelligence and Automation: This invention deeply integrates AI technology into the logistics scanning and sorting process, enabling image self-correction and automatic classification functions. It promotes the intelligent and automated development of logistics warehouses, provides strong support for the digital transformation of the logistics industry, and helps improve the competitiveness and market adaptability of logistics companies. Attached Figure Description

[0041] Figure 1 This is a structural block diagram of an intelligent scanning platform for image self-correction according to the present invention.

[0042] Figure 2 This is a flowchart of a sorting optimization method according to the present invention. Detailed Implementation

[0043] Please see Figure 1 As shown, in a first aspect of the present invention, an image self-correcting intelligent scanning platform is provided, comprising:

[0044] A multi-angle image acquisition module is configured above and to the side of the sorting line transport device to dynamically acquire image information of the items to be identified, including barcode images and images of the item's appearance.

[0045] A barcode repair and recognition module is connected to the multi-angle image acquisition module. The barcode repair module has a built-in barcode AI repair algorithm, which is used to repair blurry or damaged barcode images and identify items based on the repaired barcode.

[0046] An image classification module is connected to the multi-angle image acquisition module. The image classification module has a built-in AI classification model for item images, which is used to automatically mark the type of goods based on the barcode image and the appearance image of the item during the scanning process.

[0047] An image evaluation module is connected to the barcode repair and recognition module and the image classification module. The image evaluation module is used to identify the item type by combining the barcode recognition results and AI classification labels.

[0048] The item sorting module, connected to the image evaluation module, is used to generate sorting path instructions based on the item type and coordinate with the sorting robotic arm to execute them.

[0049] The multi-angle image acquisition module includes acquisition devices configured above and to the side of the sorting line transport device. The acquisition devices are high-definition industrial cameras equipped with ring LED fill lights with adjustable brightness to improve the lighting conditions of the goods images, especially in dimly lit warehouse environments or when the goods are highly reflective, to ensure the clarity of barcode and item appearance images.

[0050] Subsequently, basic preprocessing was performed on the image information acquired by the high-definition industrial camera. The image information included barcode images and product appearance images. The preprocessing steps included Gaussian smoothing for noise reduction and histogram equalization to enhance contrast.

[0051] The high-definition industrial camera used is the Baslerac A1920 series, with a resolution of 1920×1200, a frame rate of 30FPS, and a fixed-point acquisition mode (static frame). Images are saved in PNG format, with each pixel value ranging from 0 to 255. Fixed exposure parameters (e.g., 20ms exposure time, f / 5.6 aperture) should be used during acquisition to suppress the influence of complex lighting conditions on the sorting and transport line.

[0052] Goods in the logistics warehouse are transported to the scanning area via conveyor belt. The conveyor belt speed is set to 0.5 meters per second to ensure that the camera has enough time to capture images of the goods from multiple angles.

[0053] When goods enter the scanning area, industrial cameras installed on both sides of the conveyor automatically trigger the shooting function to acquire image information of the goods from different angles. At the same time, the ring LED supplementary light automatically adjusts its brightness according to the ambient light intensity detected by the light sensor to ensure image clarity.

[0054] Basic preprocessing is performed on images acquired by high-definition industrial cameras, including Gaussian smoothing for noise reduction and histogram equalization to enhance contrast. A recommended Gaussian filter kernel size is [size to be specified]. Standard deviation Then, histogram equalization was used to enhance the contrast.

[0055] The preprocessed image information, including barcode images and product appearance images, is then sent to the barcode repair and recognition module and the image classification module, respectively.

[0056] It adopts a high-performance industrial-grade computer, equipped with an Intel Core i7 processor, 32GB of memory and a 1TB solid-state drive, to meet the computing and storage needs of AI repair algorithms and classification models, and ensure the stable operation of the platform; it is connected to a network attached storage (NAS) device with a storage capacity of 10TB, which is used to store a large amount of scanned images, repaired barcode information, classification results and other data, so as to facilitate subsequent query, statistics and analysis.

[0057] AI repair algorithms include:

[0058] The identification unit is used to identify blurred or damaged areas in a barcode image and determine their location and extent.

[0059] The image restoration model is trained using a large number of blurry and damaged barcode images and their corresponding standard clear barcode images. Then, the trained deep learning model is used to restore the blurry or damaged barcode images, generating restored clear barcode images, and then the items are identified based on the restored clear barcode images.

[0060] For the barcode restoration algorithm, a convolutional neural network (CNN) model was trained by collecting 100,000 barcode image samples of different types (such as product barcodes, express delivery tracking numbers, etc.) with varying degrees of blurriness and damage, along with corresponding standard clear barcode images. During training, a learning rate of 0.001 was set, the training period was 100 epochs, and the cross-entropy loss function was used. By continuously adjusting the model parameters, the similarity between the restored barcode image and the standard clear barcode image reached over 95%.

[0061] The AI ​​classification model for item images employs transfer learning. Based on a pre-trained ResNet-50 model, it undergoes fine-tuning using 50,000 labeled item appearance images (representing over 20 categories such as electronics, clothing, and food) from a logistics warehouse. During training, a learning rate of 0.0001 and a training period of 50 epochs were set. The Adam optimizer was used, resulting in a classification accuracy exceeding 90%.

[0062] Finally, an image restoration accuracy index is introduced to evaluate the restoration results. The calculation formula for the image restoration accuracy index is as follows:

[0063] ,

[0064] in, Indicates the accuracy index of image restoration. Indicates the deviation coefficient. This represents the ground-corrected image with position coordinates (i, j) on the image. This represents the output image with position coordinates (i, j) on the image. Indicates Centered on, along its main migration direction The larger the deviation of the extended linear neighborhood window, the greater the fluctuation of the shadow of that point within its directional neighborhood, which does not conform to the principle of diffusion path stationarity. N represents the number of samples, an image restoration accuracy metric. The closer the value is to 0, the higher the repair accuracy.

[0065] This formula is the image-level physical stability residual function, which is inspired by the rate gradient consistency principle derived from the two-dimensional simplified diffusion equation. That is, in the main direction of diffusion of image defects or shadows, the velocity change should be gradual and there should be no abrupt changes between the velocity vectors at each point.

[0066] This formula is designed with full consideration of the specific characteristics of the image scene. Image defects or shadows should not appear abrupt or reversed, especially when guided by the missing direction; shadows should be consistent in direction and smooth in velocity. This is achieved by constructing directional neighborhoods. Instead of ordinary The region, by enhancing its response to the "missing migration main direction", avoids being misled by lateral noise interference.

[0067] For example: if the speed of a certain shadow trail is approximately 1.0, 1.1, and 1.2 (pixels / frame) between frames, then It is close to 0; however, if the path suddenly changes from 1.0 to 3.5, and then back to 1.2, then... If the elevation is significantly higher, the system can determine that the path segment is "non-stationary diffusion," thus reducing its reliability.

[0068] The other path transmits the image data of the object's appearance to the image classification module. The image classification software first extracts features from the image, using the ResNet-50 model to extract key features such as the object's shape, color, and texture.

[0069] The extracted feature vectors are input into the trained AI classification model for item images. The model classifies and identifies items based on the learned knowledge of the goods type features and assigns corresponding goods type labels, such as "electronic products - mobile phones" and "clothing - T-shirts".

[0070] The classification results are associated with the corresponding barcode information to form complete cargo identification data, which is stored in the storage device and sent to the sorting control system.

[0071] AI classification models for object images include:

[0072] The feature extraction unit is used to extract key features from product appearance images, including shape, color, and texture;

[0073] The classification training unit, including a deep learning classification model, is trained using a large number of labeled item images and can accurately identify different types of goods.

[0074] The classification execution unit is used to classify and identify newly scanned images of the appearance of items based on a trained deep learning classification model, automatically label the type of goods, generate classification result information, and store it in association with the corresponding barcode information.

[0075] After the camera acquires an image of the goods, the image data is first transmitted to the barcode repair and recognition module. This module uses professional barcode recognition software to locate and decode the barcode in the image, and outputs a confidence score for barcode recognition. The confidence score reflects the degree to which the barcode information is correctly recognized, and its calculation is based on multiple dimensions of features such as the clarity and completeness of the barcode image and its matching degree with the standard barcode template.

[0076] For a clear, undamaged barcode that conforms to standard coding rules, the confidence level of recognition may be as high as 98%; while for a barcode that is partially covered by tape and has stains, the confidence level may be only 70%.

[0077] Simultaneously, the cargo image data is also sent to the item classification module. This module uses a trained deep learning classification model to analyze the appearance of the items and outputs a predicted category and corresponding confidence score. The classification confidence score reflects the model's certainty regarding the classification result and is determined based on factors such as the feature completeness of the item image and its similarity to training samples. For example, for clear images of electronic products with obvious features, the classification confidence score can reach 95%; for items with blurry appearances and atypical features, the confidence score may drop to 60%.

[0078] When the barcode recognition confidence level is greater than 90%, it indicates that the barcode information has high reliability. At this point, the system prioritizes directly using the product information carried by the barcode (such as product number, batch number, etc.) for subsequent sorting decisions. This strategy fully utilizes the accuracy and uniqueness of barcode information, quickly identifying the product when the barcode is reliable, accelerating the sorting process, avoiding unnecessary complex calculations, and improving system efficiency. For example, a batch of brand-new smartphones that have just entered the warehouse has standardized and clear barcodes with a 95% recognition confidence level. The system immediately determines the product type and sorting destination based on the barcode information, without needing to refer to the item classification results further.

[0079] When the barcode recognition confidence level is less than or equal to 90%, it indicates that there is some uncertainty in the barcode information, which may be due to reasons such as partial obstruction, wear and tear, or blurry printing. In this case, the system activates the comprehensive decision-making mode, which weights and fuses the item classification results with the barcode information to generate a more accurate goods identification result.

[0080] The specific weighting method is as follows:

[0081] ,

[0082] in, Indicates weighted decision-making. These are weighting coefficients. For the barcode recognition results of the item, The image recognition results for the item.

[0083] and + =1. In the initial state, this can be set based on a large amount of experimental data and experience. =0.6, =0.4, emphasizing the fundamental role of barcode information while also taking into account its supplementary value in item classification.

[0084] Confidence level of barcode recognition Confidence of item classification The overall confidence level is obtained by multiplying each product by its corresponding weight and then summing the results. ,Right now:

[0085] ,

[0086] The barcode information is corrected or confirmed based on the overall confidence level and the item classification results. For example, if the confidence level for barcode recognition due to stains is 80% ( =80%), the item classification model determines that the goods are a certain brand of shampoo, with a confidence level of 90%. =90%), calculated according to the above weights =0.6×80%+0.4×90%=84%. If the overall confidence level reaches the preset threshold (e.g., 80%), the item classification result will be used as an auxiliary basis to correct or confirm the category of goods pointed to by the barcode information, ensuring the accuracy of the sorting decision; if the overall confidence level is lower than the threshold, the information of the goods will be marked as questionable, and a manual review process will be initiated, with staff making the final judgment to ensure that the sorting of goods is foolproof.

[0087] The confidence level C is calculated as follows:

[0088] ,

[0089] in, Indicates the angle of the barcode. The pose angle is a counterclockwise rotation angle based on the horizontal line of the image, and the unit is degrees. N is the number of samples.

[0090] If the angle between the barcode border and the barcode shadow changes beyond a certain threshold, it will be considered perceptually unstable, reducing the confidence level of that frame. This mechanism can effectively identify inconsistencies in barcode orientation when the barcode is affected by shadows or missing elements. The frame is automatically reduced to below a preset threshold, prompting the platform system to temporarily suspend its use.

[0091] Through this dual-channel analysis and result fusion strategy, the present invention can flexibly and intelligently process cargo information under different conditions, give full play to the advantages of barcode recognition and item classification, effectively cope with the complex and ever-changing cargo identification scenarios in logistics warehouses, further improve sorting efficiency and accuracy, reduce the sorting error rate caused by information errors, and provide solid technical support for the efficient operation of logistics warehouses.

[0092] The item sorting module includes a sorting robotic arm, which sorts items into corresponding target areas according to item type.

[0093] After receiving the goods identification data sent by the platform host, the sorting control system plans the sorting path according to the preset sorting rules and warehouse layout. For example, electronic products are sorted to the corresponding electronic equipment storage area according to the type of goods, and clothing is sorted to the clothing storage area, etc.

[0094] The sorting control system sends sorting instructions to the sorting robotic arm, sorting robot, or automated guided vehicle (AGV), including information such as the location of the goods on the conveyor line, the target storage area, and the transportation route.

[0095] Sorting robotic arms, sorting robots, or AGVs precisely grab goods from the conveyor line according to instructions and transport them to the target storage location along the planned path, completing the sorting and warehousing operation.

[0096] The intelligent scanning platform for image self-correction of the present invention also includes a data update and model optimization mechanism:

[0097] Regularly collect newly emerging barcode types and product appearance images in the logistics warehouse to expand the training sample set of the barcode repair algorithm and the training dataset of the item image classification model. This includes collecting new data once a quarter, adding 10,000 barcode images and 5,000 item appearance images each time.

[0098] The barcode repair model and classification model were retrained and optimized using the newly added data to continuously improve their generalization ability and recognition accuracy. Through continuous optimization, the accuracy of the barcode repair model in repairing new types of blurred and damaged barcodes was improved, and the accuracy of the item classification model in classifying new types of goods was increased.

[0099] A feedback mechanism and an error correction mechanism were also set up:

[0100] During the goods sorting process, if the sorting robot or staff finds a sorting error (such as incorrect goods type labeling, incorrect barcode information recognition causing goods to be sent to the wrong storage area, etc.), the error information can be promptly fed back to the image self-correcting intelligent scanning platform through a handheld terminal device.

[0101] After receiving feedback, the platform analyzes the corresponding barcode repair algorithm and image classification model to identify the causes of the errors, such as model parameter deviations or the omission of barcode damage features. To address these issues, the platform makes targeted adjustments and optimizations to the model and retrains the relevant model parameters.

[0102] The optimized model was put back into operation, and the previously erroneous cargo image data was reprocessed and identified to verify the model correction effect, ensure the accuracy of subsequent sorting operations, and continuously improve the stability and reliability of the system.

[0103] Through the above embodiments, the present invention can effectively solve the problem of identifying tape-covered and soiled barcodes in logistics warehouses, reduce the labor costs of sorting centers, and achieve efficient and accurate logistics sorting optimization. It has significant economic and social benefits and can be widely applied to various logistics and warehousing enterprises, promoting the intelligent development of the logistics industry.

[0104] Please see Figure 2 As shown, in a second aspect of the present invention, a sorting optimization method is also provided, comprising the following steps:

[0105] S1: Dynamically acquire multi-view images of items in transit using industrial high-definition cameras configured above and to the side of the sorting line transport device;

[0106] S2: Perform an integrity assessment on the barcode area. If there is any blurring or damage, repair and recognize the barcode using a barcode AI repair algorithm.

[0107] S3: Parallel execution of barcode recognition and item image classification dual-channel analysis, wherein the item image classification is achieved through an image AI classification model;

[0108] S4: Combines barcode recognition results and AI classification labels to identify the item type;

[0109] S5: Maps the recognition results to the sorting instruction set, driving the sorting robotic arm to automatically sort items.

[0110] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. An intelligent scanning platform for image self-correction, characterized in that, include: A multi-angle image acquisition module is configured above and to the side of the sorting line transport device to dynamically acquire image information of the items to be identified, including barcode images and images of the item's appearance. A barcode repair and recognition module is connected to the multi-angle image acquisition module. The barcode repair module has a built-in barcode AI repair algorithm, which is used to repair blurry or damaged barcode images and identify items based on the repaired barcode. An image classification module is connected to the multi-angle image acquisition module. The image classification module has a built-in AI classification model for item images, which is used to automatically mark the type of goods based on the barcode image and the appearance image of the item during the scanning process. An image evaluation module is connected to the barcode repair and recognition module and the image classification module. The image evaluation module is used to identify the item type by combining the barcode recognition results and AI classification labels. The item sorting module, connected to the image evaluation module, is used to generate sorting path instructions based on the item type and coordinate with the sorting robotic arm to execute them. The AI ​​repair algorithm includes: The identification unit is used to identify blurred or damaged areas in a barcode image and determine their location and extent. The image restoration model is trained using a large number of blurry and damaged barcode images and their corresponding standard clear barcode images. The trained deep learning model then restores the blurry or damaged barcode images, generating restored clear barcode images. Finally, the model identifies items based on these restored clear barcode images. The restoration results are evaluated by introducing an image restoration accuracy index, which is calculated using the following formula: , in, Indicates the accuracy index of image restoration. Indicates the deviation coefficient. This represents the ground-corrected image with position coordinates (i, j) on the image. This represents the output image with position coordinates (i, j) on the image. Indicates Centered on, along its main migration direction The extended linear neighborhood window, the greater the deviation, the greater the fluctuation of the shadow of that point within its directional neighborhood. N represents the number of samples, an image inpainting accuracy metric. The closer the value is to 0, the higher the repair accuracy. The image evaluation module identifies the type of item through the following steps: When the confidence level for barcode recognition is greater than 90%, barcode information should be used preferentially. When the confidence level of barcode recognition is less than or equal to 90%, a weighted decision is made by combining the item classification results with the barcode information; Weighted decision is calculated using the following formula: ， in, Indicates weighted decision-making. These are weighting coefficients. For the barcode recognition results of the item, The image recognition results for the item.

2. The intelligent scanning platform for image self-correction according to claim 1, characterized in that, The multi-angle image acquisition module includes acquisition devices configured above and to the side of the sorting line transport device. The acquisition devices are high-definition industrial cameras. The image information acquired by the high-definition industrial camera is then subjected to basic preprocessing. The preprocessing steps include Gaussian smoothing for noise reduction and histogram equalization to enhance contrast.

3. The intelligent scanning platform for image self-correction according to claim 1, characterized in that, The AI ​​classification model for the item images includes: The feature extraction unit is used to extract key features from product appearance images, including shape, color, and texture; The classification training unit, including a deep learning classification model, is trained using a large number of labeled item images and can accurately identify different types of goods. The classification execution unit is used to classify and identify newly scanned images of the appearance of items based on a trained deep learning classification model, automatically label the type of goods, generate classification result information, and store it in association with the corresponding barcode information.

4. The intelligent scanning platform for image self-correction according to claim 1, characterized in that, The item sorting module includes a sorting robotic arm, which sorts items into corresponding target areas according to item type.

5. A sorting optimization method based on the platform described in any one of claims 1-4, characterized in that, Includes the following steps: S1: Dynamically acquire multi-view images of items in transit using industrial high-definition cameras configured above and to the side of the sorting line transport device; S2: Perform an integrity assessment on the barcode area. If there is any blurring or damage, repair and recognize the barcode using a barcode AI repair algorithm. S3: Parallel execution of barcode recognition and item image classification dual-channel analysis, wherein the item image classification is achieved through an image AI classification model; S4: Combines barcode recognition results and AI classification labels to identify the item type; S5: Map the recognition results to the sorting instruction set to drive the sorting robotic arm to automatically sort items.

Images