Intelligent identification and reporting system for damaged express parcels

Abstract

The invention, which relates to the technical field of logistics information, discloses an intelligent identification and reporting system for damaged express parcels, comprising an automatic identification and photographing module, a deep identification and prompting module, a reporting and printing module and a data interaction module. And the automatic identification and photographing module is used for performing automatic focusing photographing on the express waybill to obtain a waybill image, and extracting a waybill number by adopting an image identification algorithm. According to the intelligent identification and reporting system for the damaged express, through intelligent processing of links of automatic identification of waybill numbers, photographing, prompt reporting and new express bill printing, the time and steps of manual operation are reduced, the processing efficiency of the damaged express is improved, the processing period is shortened, and the processing efficiency is improved. Therefore, the express company can respond to and process the problem of the damaged piece more quickly, the intervention of human factors is reduced, errors caused by manual input and error judgment are avoided, and the accuracy of identifying and reporting the damaged piece of the express is improved.

Description

Technical Field

[0001] The present invention relates to the field of logistics information technology, and particularly to an intelligent identification and reporting system for damaged express packages. Background Art

[0002] As an important part of the modern logistics system, the operation efficiency and service quality of the express delivery industry are directly related to the shopping experience of the majority of consumers. During the transportation and distribution of express deliveries, damage to express packages is an issue that cannot be ignored. Traditionally, the identification and reporting of damaged express packages mainly rely on the manual operations of couriers. This process is not only time-consuming and laborious but also difficult to ensure accuracy and timeliness. After a courier discovers a damaged express package, they need to manually enter the express waybill number, take photos of the damage, evaluate the degree of damage, and report it to the express company through specific channels. The entire process is cumbersome and inefficient;

[0003] However, the existing manual identification and reporting method has many problems and deficiencies. On the one hand, due to the large workload and tight time of couriers, the process of manual entry and photo-taking is prone to errors, resulting in inaccurate records or omission of important information. On the other hand, the subjective judgment of the degree of damage exists, and different couriers may have different evaluation results for the same damage situation. This not only affects the objectivity of the data but also brings unnecessary trouble to subsequent claims and handling work. In addition, communication barriers and information delays in the manual reporting process are also important factors affecting service quality, often leading to a decrease in customers' trust in express companies and a decline in satisfaction. In response to this, we propose an intelligent identification and reporting system for damaged express packages. Summary of the Invention

[0004] To solve the above technical problems, an intelligent identification and reporting system for damaged express packages is provided, and this technical solution solves the above problems.

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

[0006] An intelligent identification and reporting system for damaged express packages, comprising: an automatic identification and photo-taking module, a depth identification and prompting module, a reporting and printing module, and a data interaction module;

[0007] The automatic identification and photo-taking module is used to automatically focus on and photograph the express waybill to obtain a waybill image, and use an image recognition algorithm to extract the waybill number, and this module has an adaptive recognition function for the waybill formats of different express companies;

[0008] The depth recognition and prompt module is electrically connected to the automatic recognition and photographing module. When the waybill number cannot be fully recognized, the depth recognition and prompt module is used to perform in-depth matching by analyzing the remaining fields on the waybill, calculate the similarity between the remaining fields and historical waybills using a text semantic model, and combine image segmentation technology to more clearly identify the boundary of the damaged area to further improve the accuracy of damage degree assessment, evaluate the damage degree, and prompt the user to perform corresponding reporting operations according to the damage degree of the package;

[0009] The reporting and printing module is electrically connected to the depth recognition and prompt module. The reporting and printing module is used to trigger the selection of corresponding work order types according to the damage score given by the depth recognition and prompt module, report seriously damaged work orders and generate new waybills;

[0010] The data interaction module is used to communicate with the work order processing system through the RESTful API. The work order data format is JSON, including the waybill number, damaged pictures, scores, and hash values.

[0011] Preferably, the automatic recognition and photographing module includes: a camera unit, an image recognition unit, and a waybill number verification unit;

[0012] The camera unit is used to automatically focus on and photograph the express waybill to obtain the waybill image, and supports adaptive photographing of waybill formats of different express companies;

[0013] The image recognition unit is used to extract the waybill number based on the obtained waybill image and has the ability to recognize waybill formats of different express companies;

[0014] The waybill number verification unit is used to verify the validity of the extracted waybill number through an algorithm.

[0015] Preferably, the camera unit supports dynamic exposure adjustment, and the exposure parameters are calculated by the following formula:

[0016] The brightness is evaluated by calculating the average brightness of the image, and the calculation formula for the average brightness of the image is:

[0017]

[0018] In the formula, L avg represents the average brightness value of the image, W and H respectively represent the width and height of the image, ∑x,y represents the sum of the brightness values of all pixels in the image, x and y respectively represent the abscissa and ordinate of the pixel, and I gray (x, y) represents the gray value of the image at the coordinate (x, y);

[0019] The target brightness is set to 128, and the exposure time adjustment formula is:

[0020]

[0021] Where, T new Represents the adjusted exposure time, T current Indicates the current exposure time, L target Indicates target brightness;

[0022] The image recognition module supports multi-language express bill processing and recognition of express bill formats of different express companies, including:

[0023] A database containing the features of common express delivery company delivery bill templates is established. The deep learning convolutional neural network is used to extract the features of the input delivery bill image, and the features are compared and matched with the template features in the database to calculate the similarity score. When the similarity exceeds the set threshold, the express delivery company format to which the delivery bill belongs is determined. The calculation formula for the similarity score is:

[0024]

[0025] In the formula, A i and B i Represent the i-th element of vectors A and B respectively;

[0026] Use the n-gram model to calculate the language probability distribution of the text, select the language with the maximum probability, and load the corresponding OCR character set template according to the detection results;

[0027] For multi-language mail bills, a joint character set is used for matching, and the similarity is calculated as the weighted sum of the scores of each language template:

[0028]

[0029] where α k is the language weight, S k is the matching score for the kth language.

[0030] Preferably, the method steps of extracting the waybill number using an image recognition algorithm are:

[0031] The obtained face sheet image is grayed out, where the conversion formula for graying out is:

[0032] I gray =0.299R+0.587G+0,114B

[0033] Where R, G and B are the RGB channel values ​​of the original image;

[0034] Based on the grayscale processed face order image, an adaptive Gaussian filter is applied to denoise the image. The filter kernel function is:

[0035]

[0036] Wherein, G(x,y) represents the value of the Gaussian function at the point with coordinates (x,y) in the image plane, and σ represents the standard deviation of the Gaussian function. is the exponential part of the Gaussian function, and x 2 +y 2 represents the square of the Euclidean distance between a certain point and the central pixel. is the normalization coefficient;

[0037] After denoising, the OCR technology is used to locate the waybill number area, and individual characters are extracted through a character segmentation algorithm. The character segmentation adopts the vertical projection method, and the projection function is:

[0038]

[0039] Wherein, I binary is the binary image, and H is the height of the image;

[0040] Based on template matching to verify the validity of the waybill number, the cosine similarity is used for calculating the matching similarity:

[0041]

[0042] Wherein, C is the feature vector of the recognized character, and T is the feature vector of the template character.

[0043] Preferably, the depth recognition and prompt module includes: a depth recognition unit, a damage degree evaluation unit, and a damage degree prompt unit;

[0044] The depth recognition unit is used to perform depth matching by analyzing the remaining fields of the waybill when the waybill number cannot be completely recognized. The remaining fields of the waybill include, but are not limited to: name, address, phone number, and three-segment code, and this unit combines image segmentation technology to more clearly identify the boundary of the damaged area during the analysis process;

[0045] The damage degree evaluation unit is used to evaluate the damage degree of the waybill that cannot be completely recognized and output a damage score;

[0046] The damage degree prompt unit is used to prompt the user to perform corresponding reporting operations according to the damage degree of the package.

[0047] Preferably, the depth recognition unit calculates the similarity between the remaining fields and the historical waybill based on the text semantic model of BERT;

[0048] Among them, the training steps of the BERT model are:

[0049] In the pre-training stage, the historical waybill dataset of express delivery waybills is used for masked language modeling, and its loss function is:

[0050] LMLM = -∑i∈M logP(w i │W\i)

[0051] In the formula, M represents the set of mask positions;

[0052] In the fine-tuning stage, contrastive learning is used to optimize semantic embedding, and its loss function is:

[0053] L contrast = max(0, Sim(q, k - ) - Sim(q, k + ) + ∈)

[0054] In the formula, k + is the positive sample, k - is the negative sample, and ∈ is the margin threshold;

[0055] Among them, the similarity calculation formula is:

[0056]

[0057] In the formula, Sim represents similarity, Embed dam represents the residual field, and Embed his represents the historical waybill field.

[0058] Preferably, the damage degree evaluation unit adopts a multi-scale convolutional neural network. The input is the multi-angle image of the package processed by combining image segmentation technology, and the output is the damage score;

[0059] Among them, the image segmentation technology adopts a semantic segmentation model based on deep learning. By training a large number of express package images marked with damaged areas, it can accurately identify the damaged positions and ranges on the surface of the package, and standardize the segmented damaged area images into RGB three-channel images with a size of 224x224 as the input of the multi-scale convolutional neural network;

[0060] Among them, the multi-scale convolutional neural network includes the following layers:

[0061] Input layer, receiving the image processed as above;

[0062] First convolutional layer, using a 3x3 convolutional kernel, a stride of 1, and outputting a 64-channel feature map. The activation function is ReLU;

[0063] Max pooling layer, with a 2x2 pooling kernel and a stride of 2;

[0064] Second convolutional layer, with a 5x5 convolutional kernel and outputting a 128-channel feature map;

[0065] Multi-scale feature fusion layer, which fuses the feature maps of different convolutional layers through upsampling and splicing operations. The fusion formula is: F fuse = Concat(Upsample(F1), F2), where F1 and F2 are feature maps of different layers;

[0066] Fully connected layer, which outputs the damage score. The loss function is the mean squared error:

[0067]

[0068] In the formula is the true score;

[0069] Among them, the scoring formula for the damage score is:

[0070]

[0071] In the formula, D represents the damage score, w i represents the feature weight, and f i represents the i-th layer convolutional feature.

[0072] Preferably, the logic of the damage degree prompt module is:

[0073] If the damage score D < θ1, prompt "slight damage" and ask the user to confirm whether to report;

[0074] If θ1 ≤ D < θ2, prompt "moderate damage", recommend replacing the packaging and reporting;

[0075] If D ≥ θ2, prompt "severe damage", and force to replace the packaging and report;

[0076] The thresholds θ1 and θ2 are determined by training with historical data, and satisfy θ1 = μ - δ, θ2 = μ + δ, where μ is the average damage score and δ is the standard deviation.

[0077] Preferably, the reporting and printing module includes: a damaged waybill reporting unit and a new waybill printing unit;

[0078] The damaged waybill reporting unit is used to report severe damage work orders

[0079] The new waybill printing unit is used to generate a new waybill from the reported severe damage waybill. When printing the new waybill, a unique identifier is generated through the hash algorithm:

[0080] Concatenate the waybill number and the timestamp into a string and encode it as a UTF-8 byte stream;

[0081] Apply SHA-256 to generate a fixed-length hash value and embed the hash value into the QR code of the new waybill. Among them, the formula for generating a unique identifier by the hash algorithm is

[0082] Hash = SHA-256(N ub +T po )

[0083] Wherein, N ub is the waybill number, and T po is the timestamp.

[0084] Preferably, the work order reporting process of the data interaction module includes:

[0085] Encrypt the JSON data using the AES-256 algorithm, and the key derivation function is PBKDF2: key = PBKDF2(Password, Salt, Iterations), where the number of iterations ≥ 10,000;

[0086] Sign the encrypted data using the RSA algorithm, and the signature formula is: Signature = RSASign(Hash(Data), PrivateKey).

[0087] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0088] The intelligent identification and reporting system for damaged express packages proposed by the present invention reduces the time and steps of manual operations through intelligent processing of automatic identification of waybill numbers, taking pictures, prompting for reporting, and printing new waybills, improves the efficiency of handling damaged express packages, shortens the processing cycle, enables express companies to respond and handle damaged package problems faster, reduces the intervention of human factors, avoids errors caused by manual entry and judgment mistakes, improves the accuracy of identifying and reporting damaged express packages, reduces subsequent processing problems caused by incorrect information, improves the reliability of the entire processing process, quickly and accurately processes damaged express packages, can timely discover and solve express package damage problems, improves customer satisfaction, enhances customers' trust and loyalty to express companies, thereby improving the overall service quality of express companies, establishing a good brand image, realizing the automation and intelligence of the damaged express package processing process, optimizing the damaged express package processing process in the express industry, providing a more efficient and reliable solution for express companies, reducing labor costs, improving operational efficiency, and enhancing the market competitiveness of express companies. Description of the Drawings

[0089] Figure 1 is the system module framework diagram of the present invention. Detailed Embodiments

[0090] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations.

[0091] Refer toFigure 1 As shown in the figure, an intelligent identification and reporting system for damaged express packages includes: an automatic identification and photographing module, a deep identification and prompting module, a reporting and printing module, and a data interaction module;

[0092] The automatic identification and photographing module is used to automatically focus on and photograph the express waybill, obtain the waybill image, extract the waybill number by using an image recognition algorithm, and this module has the function of adaptive recognition of waybill formats of different express companies;

[0093] The deep identification and prompting module is electrically connected to the automatic identification and photographing module. When the waybill number cannot be completely recognized, the deep identification and prompting module is used to perform deep matching by analyzing the remaining fields of the waybill, calculate the similarity between the remaining fields and historical waybills by using a text semantic model, and combine image segmentation technology to more clearly identify the boundary of the damaged area to further improve the accuracy of damage degree assessment, evaluate the damage degree, and prompt the user to perform corresponding reporting operations according to the damage degree of the package;

[0094] The reporting and printing module is electrically connected to the deep identification and prompting module. The reporting and printing module is used to trigger the selection of corresponding work order types according to the damage score given by the deep identification and prompting module, report seriously damaged work orders and generate new waybills;

[0095] The data interaction module is used to communicate with the work order processing system through the RESTful API. The work order data format is JSON, including the waybill number, damaged pictures, scores, and hash values.

[0096] This intelligent identification and reporting system for damaged express packages can automatically focus on and photograph the waybill to obtain an image and extract the waybill number. When the waybill number recognition fails, it can also perform deep matching to evaluate the damage degree, trigger work order reporting according to the score and generate new waybills, and also communicate with the work order processing system through the RESTful API, greatly improving the processing efficiency and accuracy of damaged express packages and reducing labor costs.

[0097] The automatic identification and photographing module includes: a camera unit, an image recognition unit, and a waybill number verification unit;

[0098] The camera unit is used to automatically focus on and photograph the express waybill to obtain the waybill image, and supports adaptive photographing of waybill formats of different express companies;

[0099] The image recognition unit is used to extract the waybill number based on the obtained waybill image and has the ability to recognize waybill formats of different express companies;

[0100] The waybill number verification unit is used to verify the validity of the extracted waybill number through an algorithm.

[0101] The camera unit supports dynamic exposure adjustment, and the exposure parameters are calculated by the following formula:

[0102] The brightness is evaluated by calculating the average image brightness, and the formula for calculating the average image brightness is:

[0103]

[0104] In the formula, L avg represents the average brightness value of the image, W and H respectively represent the width and height of the image, ∑x,y represents the sum of the brightness values of all pixels in the image, x and y respectively represent the abscissa and ordinate of the pixel, and I gray (x, y) represents the gray value of the image at the coordinate (x, y);

[0105] The target brightness is set to 128, and the exposure time adjustment formula is:

[0106]

[0107] In the formula, T new represents the adjusted exposure time, T current represents the current exposure time, and L target represents the target brightness;

[0108] The image recognition module supports multi-language waybill processing and the recognition of waybill formats of different express delivery companies, specifically including:

[0109] Build a database containing the characteristics of common express delivery company waybill templates, use the convolutional neural network of deep learning to extract the features of the input waybill image, and compare and match them with the template features in the database to calculate the similarity score. When the similarity exceeds the set threshold, determine the express delivery company format to which the waybill belongs. The formula for calculating the similarity score is:

[0110]

[0111] In the formula, A i and B i respectively represent the i-th elements of vectors A and B;

[0112] Use the n-gram model to calculate the language probability distribution of the text, select the language corresponding to the maximum probability, and load the corresponding OCR character set template according to the detection result;

[0113] For waybills containing multiple languages, use the combined character set for matching, and the similarity calculation is the weighted sum of the scores of each language template:

[0114]

[0115] Among them, αk is the language weight, S k is the matching score of the k-th language.

[0116] The camera unit can automatically focus on the shipping label image and supports dynamic exposure adjustment. By calculating the average brightness of the image to adjust the exposure time, it can ensure clear shipping label images under different lighting conditions, providing a good foundation for subsequent recognition. The image recognition unit can extract the waybill number based on the image and also supports multi-language shipping label processing. It uses the n-gram model to determine the language, loads the corresponding OCR character set template, and performs joint character set matching on multi-language shipping labels, ensuring the accuracy and generality of recognition. The waybill number verification unit further improves the reliability of the system and the efficiency and quality of express shipping label information recognition through algorithm verification of the waybill number validity.

[0117] The method steps for extracting the waybill number using the image recognition algorithm are as follows:

[0118] Perform grayscale processing on the obtained shipping label image. Among them, the conversion formula for grayscale processing is:

[0119] I gray = 0.299R + 0.587G + 0.114B

[0120] In the formula, R, G, and B are the RGB channel values of the original image;

[0121] Based on the grayscale processed shipping label image, apply adaptive Gaussian filtering to denoise the image. The filtering kernel function is:

[0122]

[0123] In the formula, G(x,y) represents the value of the Gaussian function at the point with coordinates (x,y) in the image plane, σ represents the standard deviation of the Gaussian function, is the exponential part of the Gaussian function, x 2 + y 2 represents the square of the Euclidean distance between a certain point and the central pixel, is the normalization coefficient;

[0124] After denoising, use OCR technology to locate the waybill number area and extract individual characters through the character segmentation algorithm. The character segmentation uses the vertical projection method, and the projection function is:

[0125]

[0126] In the formula, I binary is the binary image, and H is the image height;

[0127] Verify the validity of the waybill number based on template matching. The matching similarity calculation uses cosine similarity:

[0128]

[0129] Wherein, C is the feature vector for identifying characters, and T is the feature vector of the template character.

[0130] The grayscale processing converts the color image into a grayscale image, simplifies the subsequent processing flow, and improves the processing efficiency. The adaptive Gaussian filtering can effectively remove the image noise, retain the image details, improve the image quality, and lay a foundation for accurately identifying the waybill number. Using OCR technology to locate the waybill number area and combining with the vertical projection method for character segmentation can accurately separate individual characters. Finally, based on template matching to verify the validity of the waybill number, calculating the matching degree through cosine similarity, ensuring the accuracy of the recognition result, greatly reducing the misrecognition rate, and being able to efficiently and accurately extract the waybill number from the label image to meet the requirements of rapid processing of express delivery services.

[0131] The deep recognition and prompt module includes: a deep recognition unit, a damage degree evaluation unit, and a damage degree prompt unit;

[0132] The deep recognition unit is used to perform deep matching by analyzing the remaining fields of the label when the waybill number cannot be completely recognized. The remaining fields of the label include but are not limited to: name, address, phone number, and three-segment code;

[0133] The damage degree evaluation unit is used to evaluate the damage degree of the label that cannot be completely recognized and output a damage score;

[0134] The damage degree prompt unit is used to prompt the user to perform corresponding reporting operations according to the damage degree of the package.

[0135] The deep recognition unit calculates the similarity between the remaining fields and the historical waybill based on the text semantic model of BERT;

[0136] Among them, the training steps of the BERT model are as follows:

[0137] In the pre-training stage, the historical dataset of express delivery labels is used for masked language modeling, and its loss function is:

[0138] L MLM = -∑i∈M log (w i │W\i)

[0139] Wherein, M represents the set of masked positions;

[0140] In the fine-tuning stage, contrastive learning is adopted to optimize the semantic embedding, and its loss function is:

[0141] L contrast = max(0, Sim(q, k - ) - Sim(q, k+ ) + ∈)

[0142] In the formula, k + is a positive sample, k - is a negative sample, and ∈ is the boundary threshold;

[0143] Among them, the similarity calculation formula is:

[0144]

[0145] In the formula, Sim represents similarity, Embed dam represents the residual field, and Embed his represents the historical waybill field. The deep recognition unit uses the BERT model to calculate the similarity between the residual field and the historical waybill, improving the recognition accuracy. The damage degree evaluation unit gives a score, and the prompt unit guides the user to report accordingly, forming an efficient processing process. Combining pre-training and fine-tuning to optimize the model can better handle complex situations and provide strong support for the handling of damaged express packages.

[0146] The damage degree evaluation unit adopts a multi-scale convolutional neural network. The input is the multi-angle images of the package processed by combining image segmentation technology, and the output is the damage score;

[0147] Among them, the image segmentation technology adopts a semantic segmentation model based on deep learning. By training a large number of express package images marked with damaged areas, it can accurately identify the damaged positions and ranges on the surface of the package, and standardize the segmented damaged area images, converting them into RGB three-channel images with a size of 224x224 as the input of the multi-scale convolutional neural network;

[0148] Among them, the multi-scale convolutional neural network includes the following layers:

[0149] Input layer, receiving the image processed as above;

[0150] The first convolutional layer, using a 3x3 convolutional kernel, a stride of 1, outputs a 64-channel feature map, and the activation function is ReLU;

[0151] Max pooling layer, 2x2 pooling kernel, stride 2;

[0152] The second convolutional layer, 5x5 convolutional kernel, outputs a 128-channel feature map;

[0153] Multi-scale feature fusion layer, fusing the feature maps of different convolutional layers through upsampling and splicing operations. The fusion formula is: F fuse = Concat(Upsample(F1), F2), where F1 and F2 are feature maps of different layers;

[0154] Fully connected layer, outputting the damage score, with the loss function being the mean squared error:

[0155]

[0156] In the formula is the true score;

[0157] Among them, the scoring formula for the damage score is:

[0158]

[0159] In the formula, D represents the damage score, w i represents the feature weight, and f i represents the i-th layer convolutional feature. The multi-scale convolutional neural network can process images from multiple angles, extract features through different convolutional layers, integrate information through the multi-scale feature fusion layer, and then output the damage score through the fully connected layer. Accurate calculation can make the evaluation more objective, provide a scientific basis for subsequent processing, and effectively improve the accuracy and reliability of the evaluation of damaged express packages.

[0160] The logic of the damage degree prompt module is:

[0161] If the damage score D < θ1, prompt "slightly damaged", and require the user to confirm whether to report;

[0162] If θ1 ≤ D < θ2, prompt "moderately damaged", and suggest replacing the packaging and reporting;

[0163] If D ≥ θ2, prompt "severely damaged", and force to replace the packaging and report;

[0164] The thresholds θ1 and θ2 are determined through training with historical data, satisfying θ1 = μ - δ, θ2 = μ + δ, where μ is the average damage score and δ is the standard deviation. Accurately prompt according to the damage score, allowing the user to quickly understand the damage situation. Give different treatment suggestions for different degrees, from confirming the report to forcing the replacement of the packaging and reporting, which not only meets the actual needs but also ensures the standard and orderly handling of damaged packages. Determining the thresholds based on historical data also ensures the scientific nature of the judgment.

[0165] The reporting and printing module includes: a damaged waybill reporting unit and a new waybill printing unit;

[0166] The damaged waybill reporting unit is used to report severe damage work orders

[0167] The new waybill printing unit is used to generate a new waybill from the reported severely damaged waybill. When printing the new waybill, a unique identifier is generated through the hash algorithm:

[0168] Concatenate the waybill number and the timestamp into a string and encode it as a UTF-8 byte stream;

[0169] Apply SHA-256 to generate a fixed-length hash value and embed the hash value into the QR code of the new shipping label. The formula for the hash algorithm to generate a unique identifier is

[0170] Hash = SHA-256(N ub +T po )

[0171] In the formula, N ub is the waybill number, and T po is the timestamp.

[0172] The damaged shipping label reporting unit promptly reports severely damaged work orders to ensure that problems are handled in a timely manner. The new shipping label printing unit generates a new shipping label, generates a unique identifier through the hash algorithm, combines the waybill number and the timestamp, and embeds them in the QR code to ensure the accuracy, uniqueness, and security of the shipping label information, facilitating subsequent logistics tracking and management.

[0173] The work order reporting process of the data interaction module includes:

[0174] Use the AES-256 algorithm to encrypt JSON data. The key derivation function is PBKDF2: key = PBKDF2(Password, Salt, Iterations), and the number of iterations ≥ 10000;

[0175] Use the RSA algorithm to sign the encrypted data. The signature formula is: Signature = RSASign(Hash(Date), PrivateKey). Using the AES-256 algorithm to encrypt JSON data and combining the PBKDF2 key derivation function enhances the confidentiality and anti-cracking ability of the data. Then use the RSA algorithm to sign the encrypted data to ensure the authenticity and integrity of the data source, effectively preventing data tampering, and providing a secure and reliable guarantee for the transmission of express damaged piece work order data.

[0176] Specific case analysis:

[0177] When courier Zhang was about to deliver a courier, he found that a courier package was damaged. He placed the courier package in front of the mobile phone camera, and the camera automatically focused and took a photo of the shipping label. The system identified the waybill number on the shipping label through image recognition technology.

[0178] Subsequently, according to the degree of damage, the system prompted Zhang that the courier was severely damaged and recommended reporting a severely damaged work order. Zhang clicked the severely damaged report button according to the system prompt. The system uploaded data such as the damaged package picture, waybill number, and whether to change the packaging to the work order processing system, completing the work order reporting.

[0179] After the successful reporting, the system automatically triggers the function of printing new waybills. Based on the recognized waybill number information, the system generates a new waybill, which is automatically printed by the courier's printer and repackaged.

[0180] Xiaozhang pastes the new waybill on the new package to ensure the normal circulation and subsequent processing of the express delivery.

[0181] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection required by the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent identification and reporting system for damaged express parcels, characterized in that, Including: An automatic recognition and photographing module, a depth recognition and prompting module, a reporting and printing module, and a data interaction module; The automatic recognition and photographing module is used to automatically focus on and photograph the express waybill, obtain the waybill image, and extract the waybill number by using an image recognition algorithm. And this module has the function of adaptive recognition of waybill formats of different express companies; The depth recognition and prompting module is electrically connected to the automatic recognition and photographing module. When the waybill number cannot be completely recognized, the depth recognition and prompting module is used to perform in-depth matching by analyzing the remaining fields of the waybill, calculate the similarity between the remaining fields and the historical waybill by using a text semantic model, and combine image segmentation technology to more clearly identify the boundary of the damaged area to further improve the accuracy of the damage degree assessment. According to the damage degree of the package, prompt the user to perform corresponding reporting operations; The reporting and printing module is electrically connected to the depth recognition and prompting module. The reporting and printing module is used to trigger the selection of the corresponding work order type according to the damage score given by the depth recognition and prompting module, report a seriously damaged work order and generate a new waybill; The data interaction module is used to communicate with the work order processing system through the RESTful API. The work order data format is JSON, including the waybill number, damaged pictures, scores, and hash values.

2. The intelligent recognition and reporting system for damaged express packages according to claim 1, wherein The automatic recognition and photographing module includes: a camera unit, an image recognition unit, and a waybill number verification unit; The camera unit is used to automatically focus on and photograph the express waybill to obtain the waybill image, and supports adaptive photographing of waybill formats of different express companies; The image recognition unit is used to extract the waybill number based on the obtained waybill image and has the ability to recognize waybill formats of different express companies; The waybill number verification unit is used to verify the validity of the extracted waybill number through an algorithm.

3. The intelligent recognition and reporting system for damaged express packages according to claim 2, wherein, The camera unit supports dynamic exposure adjustment, and the exposure parameters are calculated by the following formula: Evaluate the brightness by calculating the average image brightness, where the calculation formula for the average image brightness is: where L avg represents the average luminance value of the image, W and H respectively represent the width and height of the image, ∑x,y represents the summation of the luminance values of all pixels in the image, x and y respectively represent the abscissa and ordinate of the pixel, and I gray (x, y) represents the gray value of the image at the coordinate (x, y); The target brightness is set to 128, and the exposure time adjustment formula is: Where, T new represents the adjusted exposure time, T current represents the current exposure time, L target represents the target brightness; The image recognition module supports multi-language waybill processing and the recognition of waybill formats of different express companies. Specifically, it includes: Establish a database containing the template features of common express company waybills, use the convolutional neural network of deep learning to extract the features of the input waybill image, and compare and match them with the template features in the database to calculate the similarity score. When the similarity exceeds the set threshold, determine the express company format to which the waybill belongs. The calculation formula for the similarity score is: where A i and B i represent the i-th elements of vectors A and B, respectively; Use the n-gram model to calculate the language probability distribution of the text, select the language corresponding to the maximum probability, and load the corresponding OCR character set template according to the detection result; For waybills containing multiple languages, use a combined character set for matching, and the similarity calculation is the weighted sum of the scores of each language template; where α k is the language weight, and S k is the matching score of the k-th language.

4. The intelligent identification and reporting system for damaged express parcels according to claim 1, characterized in that: The method steps for extracting the waybill number by using the image recognition algorithm are: Perform grayscale processing on the obtained waybill image. Among them, the conversion formula for grayscale processing is: I gray = 0.299R + 0.587G + 0.114B In the formula, R, G, and B are the RGB channel values of the original image; Based on the grayscale processed waybill image, apply adaptive Gaussian filtering to denoise the image screen. The filtering kernel function is: Wherein, G(x, y) represents the value of the Gaussian function at the point with coordinates (x, y) in the image plane, σ represents the standard deviation of the Gaussian function, is the exponential part of the Gaussian function, x 2 +y 2 represents the square of the Euclidean distance between a certain point and the central pixel, is the normalization coefficient; After denoising, use OCR technology to locate the waybill number area, and extract individual characters through a character segmentation algorithm. The character segmentation adopts the vertical projection method, and the projection function is: where I binary is the binarized image and H is the height of the image; Verify the validity of the waybill number based on template matching. The matching similarity calculation uses cosine similarity: In the formula, C is the feature vector of the recognized character, and T is the feature vector of the template character.

5. An intelligent recognition and reporting system for damaged express packages according to claim 1, characterized in that, The deep recognition and prompt module includes: a deep recognition unit, a damage degree evaluation unit, and a damage degree prompt unit; The deep recognition unit is used to perform deep matching by analyzing the remaining fields of the waybill when the waybill number cannot be fully recognized. The remaining fields of the waybill include but are not limited to: name, address, phone number, and three-segment code. And this unit combines image segmentation technology during the analysis process to more clearly identify the boundary of the damaged area; The damage degree evaluation unit is used to evaluate the damage degree of the waybill that cannot be fully recognized and output a damage score; The damage degree prompt unit is used to prompt the user to perform corresponding reporting operations according to the damage degree of the package.

6. The intelligent recognition and reporting system for damaged express packages according to claim 5, characterized in that, The deep recognition unit calculates the similarity between the remaining fields and the historical waybill based on the text semantic model of BERT; Among them, the training steps of the BERT model are: In the pre-training stage, use the historical dataset of express waybills for masked language modeling, and its loss function is: L MLM = -∑i∈M logP(w i │W\i) In the formula, M represents the set of masked positions; In the fine-tuning stage, use contrastive learning to optimize the semantic embedding, and its loss function is: L contrast = max(0, Sim(q, k - ) - Sim(q, k + ) + ∈) where k + is a positive sample, k - is a negative sample, and ∈ is the boundary threshold; Among them, the similarity calculation formula is: Where Sim represents the similarity, Embed dam represents the residual field, Embed his represents the historical waybill field.

7. The intelligent recognition and reporting system for damaged express parcels according to claim 5, characterized in that, The damage degree evaluation unit adopts a multi-scale convolutional neural network. The input is the multi-angle image of the package processed by combining image segmentation technology, and the output is the damage score; Among them, the image segmentation technology adopts a semantic segmentation model based on deep learning. Through training on a large number of express package images marked with damaged areas, it can accurately identify the damaged positions and ranges on the surface of the package, and standardize the segmented damaged area images, converting them into RGB three-channel images with a size of 224x224 as the input of the multi-scale convolutional neural network; Among them, the multi-scale convolutional neural network includes the following layers: Input layer, receiving the image processed as above; The first convolutional layer, using a 3x3 convolutional kernel, a stride of 1, outputs a 64-channel feature map, and the activation function is ReLU; Max pooling layer, with a 2x2 pooling kernel and a stride of 2; The second convolutional layer, with a 5x5 convolutional kernel, outputs a 128-channel feature map; Multi-scale feature fusion layer, which fuses feature maps from different convolutional layers through upsampling and concatenation operations. The fusion formula is: F fuse = Concat(Upsample(F1), F2), where F1 and F2 are feature maps from different layers; Fully connected layer, outputting the damage score, and the loss function is mean squared error: where is the true score; Among them, the scoring formula for the damage score is: where D represents the damage score, w i represents the feature weight, f i represents the convolutional feature of the i-th layer.

8. The intelligent recognition and reporting system for damaged express parcels according to claim 5, characterized in that, The logic of the damage degree prompt module is: If the damage score D < θ1, prompt "Slight damage", and require the user to confirm whether to report; If θ1 ≤ D < θ2, prompt "Moderate damage", and recommend replacing the packaging and reporting; If D ≥ θ2, prompt "Severe damage", and force the replacement of the packaging and reporting; The thresholds θ1 and θ2 are determined through training with historical data, satisfying θ1 = μ - δ, θ2 = μ + δ, where μ is the average damage score and δ is the standard deviation.

9. The intelligent identification and reporting system for damaged express packages according to claim 1, wherein, The reporting and printing module includes: a damaged waybill reporting unit and a new waybill printing unit; The damaged waybill reporting unit is used to report severely damaged work orders. The new waybill printing unit is used to generate a new waybill from the reported severely damaged waybill. When printing the new waybill, a unique identifier is generated through the hash algorithm: Concatenate the waybill number and the timestamp into a string and encode it into a UTF-8 byte stream. Apply SHA-256 to generate a fixed-length hash value and embed the hash value into the QR code of the new shipping label. The formula for the hash algorithm to generate a unique identifier is Hash = SHA-256(N ub +T po ) Where N ub is the waybill number, and T po is the timestamp.

10. The intelligent identification and reporting system for damaged express parcels according to claim 1, characterized in that, The work order reporting process of the data interaction module includes: Encrypt the JSON data using the AES-256 algorithm. The key derivation function is PBKDF2: key = PBKDF2(Password, Salt, Iterations), and the number of iterations ≥ 10000. Sign the encrypted data using the RSA algorithm. The signature formula is: Signature = RSASign(Hash(Data), PrivateKey).

Images