Electromechanical product solid waste attribute intelligent identification method and system

Abstract

The invention discloses an intelligent identification method and system for solid waste attributes of mechanical and electrical products. The method comprises the following steps: automatically identifying document information through an OCR (Optical Character Recognition) technology; the method comprises the following steps: acquiring a cargo photo shot by image acquisition equipment, and acquiring an original record in a field inspection process; checking the document information with the shot goods photo and the text information of the original record; if yes, according to the determined cargo type or range, the document information is matched with key features extracted from a shot picture in a preset database, and preliminary judgment and risk assessment are carried out on the solid waste attribute of the cargo; when the goods are consistent, inputting a shot goods picture into the defect detection model, and outputting a detection result; when the two are inconsistent, receiving an artificial identification result; and taking a consistent identification result or a received artificial identification result as a final solid waste attribute identification result. According to the method, the identification efficiency and accuracy are effectively improved, and the labor cost and subjective errors are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of identification of the solid waste attributes of electromechanical products, and more particularly to an intelligent identification method and system for the solid waste attributes of electromechanical products. Background Art

[0002] At present, there are significant differences between the identification of the solid waste attributes of electromechanical products and the identification of raw material attributes. It is impossible to use instruments and equipment to inspect and analyze the composition of raw materials and use the test results to determine whether they belong to waste. The scope of electromechanical products is wide and the types are numerous. Some weigh up to several tons, and it is impossible to conduct power-on tests on each batch of products. A large amount of inspection work requires technicians to go to the inspection site to judge whether they have the original functions or uses based on the packaging, appearance, production date, safe service life, nameplate, instruction manual, and situation description of the goods. This places very high requirements on the inspection experience of inspectors and their familiarity with products. Different personnel or even different institutions are very likely to have different inspection conclusions for the same batch of goods.

[0003] The existing identification of the solid waste attributes of electromechanical products mainly relies on manual experience and has the following problems:

[0004] Low efficiency: Manual inspection is time-consuming and laborious, and it is difficult to meet the inspection needs of a large number of goods.

[0005] Poor consistency: Different inspectors may draw different conclusions for the same goods, affecting the accuracy of the identification results.

[0006] High retest risk: Manual judgment has subjective factors and is prone to errors, resulting in an increased risk of retest. Summary of the Invention

[0007] In view of this, the present invention provides an intelligent identification method and system for the solid waste attributes of electromechanical products, which are used to solve the problems in the prior art that the inspection conclusion highly depends on on-site manual identification, with high labor costs, low inspection efficiency, poor consistency, and high retest risk.

[0008] To achieve the above object, the present invention adopts the following technical solutions:

[0009] In the first aspect, an embodiment of the present invention provides an intelligent identification method for the solid waste attributes of electromechanical products, including the following steps:

[0010] S1. Automatically identify the document information through OCR technology, including: name of goods, model, quantity, declared value;

[0011] S2. Obtain the photos of the goods taken by the image acquisition device, including: photos of the transport packaging, photos of the unpacking inspection, photos of the identification inspection, and photos of the key point verification; and obtain the original records during the on-site inspection process;

[0012] S3. Check the document information against the text information extracted from the photographed goods and the originally recorded text information.

[0013] S4. When the check is consistent, match the document information with the key features extracted from the photographed goods according to the determined type or scope of the goods in a preset database, and make a preliminary judgment and risk assessment of the solid waste attribute of the goods based on the similarity.

[0014] S5. When the check is consistent, input the photographed goods photo into the defect detection model and output the detection result.

[0015] S6. When the preliminary judgment risk assessment result obtained from the preset data matching is inconsistent with the detection result of the defect detection model, receive the manual identification result; use the consistent identification result or the received manual identification result as the final solid waste attribute identification result.

[0016] Further, the original record obtained in step S2 includes:

[0017] According to the result of on-site visual inspection, make a voice description record at the same time, use a voice-to-text tool to record on-site and automatically form a text file as the original record.

[0018] Further, step S3 includes:

[0019] For some key information, use an exact matching algorithm, and determine that the match is successful when the information is exactly the same; the part of the key information includes: packing number, model and serial number on the goods nameplate, manufacturer, quantity.

[0020] For some other information, use a fuzzy matching algorithm for matching; the other part of the information includes: goods name, manufacturer name, importer name, and transportation method.

[0021] Further, step S3 also includes:

[0022] When there is a complete mismatch, send an alarm prompt to the inspector and provide the difference information; save the relevant check records and original data.

[0023] Further, step S4 includes:

[0024] When the check is consistent, according to the determined type or scope of the goods, match the document information with the key features extracted from the photographed goods in a preset database using the match library function algorithm of Pandas and match with the stored data.

[0025] When the matching similarity meets the threshold, mark the goods as potential solid waste; and based on the reasons why various products are identified as solid waste extracted from the preset database, make a preliminary judgment and risk assessment on the solid waste attributes of the goods.

[0026] Further, the preset data in step S4 includes a case base and a knowledge base;

[0027] Among them, construct a case base by collecting solid waste identification cases of electromechanical products over the years; construct a knowledge base by analyzing the case base and extracting the characteristics of solid waste.

[0028] Further, in step S5, the defect detection model is a defect detection and recognition algorithm model improved based on YOLOv8, including a backbone network, a neck network, and a head network;

[0029] Among them, in the backbone network, replace the original CSPDarknet53 network with MobileNetV3; and introduce a CA attention module on the basis of MobileNetV3; the CA attention module enhances the model's perception ability of key features by focusing on the relationship between channels.

[0030] In a second aspect, an intelligent identification system for the solid waste attributes of electromechanical products according to an embodiment of the present invention includes:

[0031] An OCR recognition module for automatically recognizing document information through OCR technology, including: goods name, model, quantity, declared value;

[0032] An information collection module for obtaining photos of goods taken by an image acquisition device, including: photos of transportation packaging, photos of unpacking inspection, photos of logo inspection, and photos of key point verification; and obtaining the original records during the on-site inspection process;

[0033] A verification module for verifying the document information with the text information extracted from the photos of the photographed goods and the text information of the original records;

[0034] A preliminary judgment module for, when the verification is consistent, matching the document information with the key features extracted from the photographed and obtained photos in the preset database according to the determined goods type or scope, and making a preliminary judgment and risk assessment on the solid waste attributes of the goods according to the similarity;

[0035] A model detection module for, when the verification is consistent, inputting the photos of the photographed goods into the defect detection model and outputting the detection result;

[0036] The identification module is used to receive the manual identification result when the preliminary judgment risk assessment result obtained by matching the preset data is inconsistent with the detection result of the defect detection model; and use the consistent identification result of the two or the received manual identification result as the final solid waste attribute identification result.

[0037] Further, it also includes:

[0038] The reference module is used to automatically prompt the historical detection information as a reference when the same goods are uploaded again next time; it includes time, consignor information, goods information, import date, inspection date, batch quantity, typical photos, identification conclusion and judgment basis.

[0039] It can be seen from the above technical solutions that compared with the prior art, the present invention has the following technical advantages:

[0040] Based on the constructed database and model, this method automatically identifies the document information, analyzes the on-site inspection pictures, and combines historical data for intelligent judgment, realizing automated information collection and intelligent identification, greatly improving the identification efficiency; based on big data and machine learning technologies, the identification results are more accurate and reliable; in addition, it also reduces the subjective factors of manual judgment and lowers the risk of re-inspection. It provides a reference basis for inspectors to assist them in making decisions. Description of the Drawings

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the provided drawings.

[0042] Figure 1 It is the flowchart of the intelligent identification method for the solid waste attribute of electromechanical products provided by the present invention.

[0043] Figure 2 It is the structural diagram of the improved YOLOv8 network provided by the present invention.

[0044] Figure 3 It is the structural block diagram of the intelligent identification system for the solid waste attribute of electromechanical products provided by the present invention. Detailed Embodiments

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0046] First, some terms related to the present invention are explained as follows:

[0047] 1. Electromechanical products: refer to various agricultural machinery, electrical appliances, and production equipment and household appliances with mechanical, electrical, and electronic properties produced using machinery, electrical appliances, and electronic equipment. Generally, it includes mechanical equipment, electrical equipment, transportation tools, electronic products, electrical appliances, instruments and meters, metal products, etc., as well as their parts and components.

[0048] 2. Solid waste: refers to solid, semi-solid, and gaseous articles and substances that have lost their original use value or have not lost their use value but are discarded or abandoned during production, life, and other activities, as well as articles and substances specified by laws and administrative regulations to be included in solid waste management. Exceptions are those that have undergone harmless processing and meet the mandatory national product quality standards, do not pose hazards to public health and ecological safety, or are determined not to be solid waste according to the solid waste identification standards and identification procedures.

[0049] 3. Identification of solid waste attributes: activities to determine whether an article or substance belongs to solid waste and to determine the type of solid waste it belongs to.

[0050] Refer to Figure 1 As shown, an intelligent identification method for the solid waste attributes of electromechanical products according to an embodiment of the present invention includes the following steps:

[0051] S1. Automatically identify document information through OCR technology, including: name of goods, model, quantity, declared value; and also including, for example: domestic consignee, import date, declaration date, etc.

[0052] S2. Obtain photos of goods taken by an image acquisition device, including: photos of transport packaging, photos of unpacking inspection, photos of label inspection, and photos of key point verification; and obtain the original records during the on-site inspection process;

[0053] Upload of on-site inspection information: Upload on-site inspection information by taking photos, including information such as transport packaging, unpacking inspection, individual packaging, retail packaging, dust-proof packaging, appearance inspection, label inspection, and key point verification.

[0054] Automatic generation of original records: Use a voice-to-text tool to automatically convert the voice descriptions of on-site inspection personnel into text files, save them, and upload them.

[0055] S3. Check the document information against the text information extracted from the photos of the goods taken and the originally recorded text information; according to the types of goods to be identified, extract key features from the document information and on-site inspection information, including packaging type, appearance status, logo integrity, usage marks on key parts, etc.; that is, compare the document information with the features extracted from the corresponding pictures.

[0056] S4. When the checks are consistent, according to the determined types or scope of goods, match the document information and the key features extracted from the photos taken, in a preset database, and based on the similarity, make a preliminary judgment and risk assessment of the solid waste attributes of the goods.

[0057] Among them, the preset database is created by collecting the solid waste identification information of port electromechanical products in recent years, creating a case library and a knowledge base, refining the reasons for various products to be identified as solid waste, and setting risk levels. Then, compare the extracted features of the goods with the data in the case library and the knowledge base, use a matching algorithm, and make a preliminary judgment and risk assessment of the goods based on the feature similarity. The risk level of the goods can also be dynamically adjusted according to the risk level set in the case library and the matching degree between the goods and the features of high-risk goods.

[0058] S5. When the checks are consistent, input the photos of the goods taken into a defect detection model and output the detection results; in this step, an improved YOLOv8 algorithm model can be used to detect and identify defects in the uploaded photo information. Evaluate whether the goods meet the solid waste attributes according to the number and severity of the defects. The improved YOLOv8 algorithm model in this step can be trained and established by collecting historical data, and learn according to historical data and manual judgment results to continuously optimize the model.

[0059] S6. When the preliminary judgment risk assessment results obtained from the preset data matching are inconsistent with the detection results of the defect detection model, receive the manual identification results; take the consistent identification results of the two or the received manual identification results as the final solid waste attribute identification results.

[0060] Based on the comprehensive database matching results, image recognition results and manual judgment results, finally determine the solid waste attributes of the goods. For example, when the manual judgment results are inconsistent with the system judgment results, take the manual judgment results as the standard and input the judgment basis. Save the current judgment results and basis into the case library for reference in the identification of similar goods next time.

[0061] This invention can improve the information input efficiency and accuracy, improve the effectiveness of the inspection plan, and enhance the on-site inspection efficiency; reduce the inspection risk; and enhance the identification accuracy. It has good practical application prospects in optimizing the inspection process, shortening the detection cycle, and improving work efficiency.

[0062] The method of the present invention will be further described in detail through a complete embodiment as follows:

[0063] In specific implementation, for example, the method can be in the form of a software program to form an intelligent identification system for the solid waste attributes of electromechanical products. The steps are as follows:

[0064] Step 1:

[0065] Determine the scope of application of the method: Electromechanical products have a wide range. Combining with the types of goods declared at the current port, the scope of objects is defined as plastic casings of laptop computers, switching power supplies, diodes, printer cartridges, liquid crystal displays, regulated power supplies, cameras, flashlights, mobile phone motherboards, memory modules, capacitors, input / output modules, mobile phone (tablet) displays, mechanical hard drives, computer motherboards, computer hosts, adapter cards, etc. Moreover, according to the increasing annual import volume of electronic components, the scope of application can be adjusted at any time.

[0066] Step 2:

[0067] Automatically identify document information: Through OCR technology, the paper version of the document is scanned and uploaded to the "Intelligent Identification System for the Solid Waste Attributes of Electromechanical Products", and is automatically identified according to different fields through OCR technology, such as the consignee within the territory, import date, declaration date, overseas shipper, mode of transport, supervision mode, etc.

[0068] Step 3:

[0069] Take pictures of on-site inspection information and upload:

[0070] Transport packaging: Simulate the packaging when taking out of the container and placing it at the inspection location. Generally, there are wooden box packaging, wooden pallets (with paper packaging boxes placed on them), and focus on whether the transport packaging is damaged or deformed, and whether there is identification information on the outside of the packaging box (such as packing list, box number, goods name, etc.).

[0071] Unpacking inspection: Check the status of the goods in the box, and focus on whether the goods have independent packaging or retail packaging, the stacking state of the goods (isolation and positioning measures), and whether there is a situation of mixed loading of different models (part numbers). If there is independent packaging or retail packaging, then check the integrity of the packaging, whether it is damaged or deformed. For certain specific goods such as chips, they need to have independent dust-proof packaging. If the goods in the box are exposed and stacked chaotically without isolation measures, it is likely to be judged as solid waste.

[0072] Independent packaging: Each piece of goods has independent packaging, and there is no situation where multiple pieces of goods share one packaging box. On-site inspection is carried out through taking pictures and visual inspection.

[0073] Retail packaging: The formal packaging that the product carries when it is sold in the circulation field, with relevant product name, manufacturer, and seller information.

[0074] Dust-proof packaging: A sealed package with a seal, commonly used for the outer packaging of precision electronic components to prevent static electricity and dust from affecting the functions of the devices.

[0075] Appearance inspection: Check whether the appearance of the goods is intact, whether there are obvious scratches, cracks, or damages on the outer shell; whether there is a crack in the liquid crystal plane; whether the buckle bracket is broken; whether the pins of the connection port are bent or deformed, etc. If there are obvious deformations, damages, or missing key components on the appearance of the goods, it is likely to be judged as solid waste.

[0076] Label inspection: Whether the nameplate label is clearly visible, whether the model and part number are clearly visible, and whether the key parameter information is complete.

[0077] Key point verification: Different products have different requirements. Focus on verifying the parts that are prone to leaving usage marks and the parts that affect key functions. Take the plastic shell of a laptop computer as an example. Check whether the threads of the panel screw holes are damaged or rusted; whether the internal buckle brackets are broken; whether the cable interfaces are intact; whether the hinges are intact and not rusted, etc.

[0078] During the shooting process, mainly through visual inspection, which is related to the familiarity of the inspectors with the products. Focus on verifying details for each different product and record them in the form of photos. Take a wireless network card as an example. Check whether there are any marks left due to plugging and unplugging on the surface of the gold fingers; whether there are any assembly marks left after screwing the screws in the surrounding screw holes; whether there are any oxidation marks left after connection on the connection socket; whether there are any marks of re-soldering on the component solder joints on the circuit board. The existence of marks indicates that the product has been used and is not a brand-new item.

[0079] Step 4:

[0080] Automatic formation of original records: The inspection site for solid waste of electromechanical products is generally a certain wharf, warehouse, etc. The environment and on-site conditions are poor, and it is impossible to record the original records methodically with a pen like in a laboratory. Therefore, existing voice-to-text tools on the market are used. According to the results of on-site visual inspection, voice descriptions are recorded simultaneously. The voice-to-text tool records the voice on-site and automatically forms a text file, which is saved to the system of the "Intelligent Identification Method for the Solid Waste Attribute of Electromechanical Products", and the original records are directly printed out later.

[0081] Step 5:

[0082] Collect the information on the identification of solid waste of electromechanical products at Shanghai Port in recent years, and create a case library and a knowledge base. Extract the reasons why a certain product is identified as solid waste. Generally, there are: 1. No necessary packaging, and there is a situation of mixed loading; 2. Obvious usage marks or damages on the appearance; 3. Incomplete structure, lacking necessary components; 4. The initial function cannot be maintained.

[0083] Set the risk level according to the frequency of being identified as solid waste:

[0084] Low risk: Low import frequency and extremely low probability of being identified as solid waste historically.

[0085] Medium risk: Medium import frequency or medium probability of being identified as solid waste historically.

[0086] High risk: High import frequency and high probability of being identified as solid waste historically.

[0087] Extremely high risk: Extremely high import frequency and extremely high probability of being identified as solid waste historically.

[0088] For example, list the goods with high import frequency and high frequency of being identified as solid waste as "high-risk goods", which mainly include the following parts:

[0089] 1) Goods feature extraction:

[0090] Starting from the types and scopes of goods determined in Step 1, for each batch of goods to be identified, the system first extracts key features according to the documentary information in Step 2 and various details in the on-site inspection information in Step 3 (such as packaging conditions, appearance states, label integrity, key point verification results, etc.). These features include but are not limited to the packaging type of the goods, whether there are damages and deformations, whether there are independent or retail packages, the scratch and cracking conditions of the appearance, the clarity and integrity of the nameplate label, the usage traces of key parts, etc.

[0091] Taking the plastic shell of a laptop as an example, the system will extract feature information such as the thread state of the panel screw holes, the internal buckle bracket situation, the integrity of the cable interface, and the hinge state, and convert it into a standardized data format for comparison with the information in the database.

[0092] 2) Database matching algorithm:

[0093] Adopt the match matching library function algorithm of Pandas to compare the extracted goods features with the existing data in the case library and knowledge base in Step 5. The algorithm will assign corresponding weights according to the importance of different features. For example, for electronic products, the integrity and usage traces of key components will be assigned higher weights, while for some shell products, the weight of appearance damage is relatively high.

[0094] When the similarity between the characteristics of the goods and those of a case identified as solid waste in the database reaches a certain threshold, the system will mark the goods as potential solid waste and prompt the relevant inspectors to further verify. At the same time, the system will also refer to the reasons why various products are identified as solid waste extracted from the knowledge base, such as unnecessary packaging, obvious signs of use on the appearance, incomplete structure, inability to maintain the initial function, etc., to make a preliminary judgment and risk assessment on the solid waste attribute of the goods.

[0095] 3) Risk level adjustment:

[0096] According to the risk level set in Step 5 and combined with the matching degree between the goods and the characteristics of high-risk goods in the database, the system will dynamically adjust the risk level of this batch of goods. If the characteristics of the goods are highly similar to those of high-risk goods, the system will correspondingly increase its risk level and give more attention and stricter inspection standards in the subsequent identification process.

[0097] Step 6:

[0098] Automatic information verification: The document information automatically identified in Step 2 and the photo information taken on-site in Steps 3 and 4 (such as container number, goods nameplate, manufacturer, importer, etc.) are uploaded to the "Intelligent Identification Method for the Solid Waste Attribute of Mechanical and Electrical Products" system for automatic verification. Consistent information will automatically turn green, and inconsistent information will turn red. The specific steps include:

[0099] 1) Information upload and integration:

[0100] In Step 6, first, the document information automatically identified in Step 2 (including domestic consignee, import date, declaration date, overseas shipper, transportation mode, supervision mode, etc.) and the photo information taken on-site in Steps 3 and 4 (such as container number, goods nameplate, manufacturer, importer, etc.) are uploaded to the "Intelligent Identification Method for the Solid Waste Attribute of Mechanical and Electrical Products" system through a secure data transmission channel.

[0101] The system will integrate and preprocess these uploaded information, extract and identify the text data in the document information and the text information in the photos, and convert them into a unified data format for subsequent verification operations.

[0102] 2) Information verification method and logic:

[0103] The system uses a combination of exact matching and fuzzy matching methods for information verification. For some key information, such as container number, model and serial number on the goods nameplate, manufacturer, etc., an exact matching algorithm is used, and it is required that the information be exactly the same to be judged as a successful match. Once these information are successfully matched, they will be displayed in green at the corresponding position in the system interface.

[0104] For some information that may have certain differences, such as the name of the importer may have differences between the abbreviated form and the full form, the expression of the transportation method may be slightly different, etc., the system adopts a fuzzy matching algorithm. By setting a certain error tolerance rate and semantic analysis rules, these information are intelligently compared. If the information is successfully matched within a reasonable error tolerance range, it will also be marked as green; if it exceeds the error tolerance range or cannot be matched, it will be marked as red, and at the same time a detailed difference report will be generated, pointing out the specific inconsistent information and possible reasons for the inspection personnel to further verify and process.

[0105] 3) Abnormal situation handling and feedback:

[0106] When there is a situation of inconsistent information (red mark), the system will automatically trigger the abnormal situation handling process. On the one hand, the system will send an instant notification to the inspection personnel, prompting the problem of inconsistent information and providing detailed difference information; on the other hand, the system will automatically save the relevant verification records and original data for subsequent traceability and analysis.

[0107] The inspection personnel can re-verify the on-site goods according to the abnormal information prompted by the system to check whether there are problems such as incorrect information entry, or the actual situation of the goods does not match the document information. If it is found that it is an incorrect information entry, the inspection personnel can correct the relevant information in the system and re-perform the verification operation; if there is a situation where the actual goods do not match the document information, it is necessary to further investigate the reasons, which may involve issues such as the source of the goods and the authenticity of the declaration, and need to be processed in accordance with relevant regulations and procedures, and the processing results are fed back to the system to improve the database and identification process.

[0108] Step 7:

[0109] Establish a judgment model based on the image for intelligent identification: According to the photos uploaded in Step 3, and in accordance with the four key points of judgment in Step 5, focus on whether there is damage, deformation, or missing, and formulate an algorithm model (when the damaged state is greater than or equal to a certain percentage of the original size, it can be judged as solid waste), and combine historical data to intelligently output the judgment result: belonging to solid waste or not belonging to solid waste.

[0110] Refer to Figure 2 As shown, establish a defect detection and recognition algorithm model - YOLOv8-KK improved based on YOLOv8:

[0111] 1. Data collection and preprocessing

[0112] Data collection: Collect a large number of pictures of electromechanical products, including normal products and solid waste products, to ensure the diversity and representativeness of the data.

[0113] Data annotation: Accurately annotate the collected images, including the categories of objects (scratches, damages, pits, etc.) and bounding boxes, to provide accurate labels for model training.

[0114] Data augmentation: Augment the images by methods such as rotating the images, scaling, adding noise, enhancing brightness, enhancing contrast, and converting to grayscale to increase the number of the dataset, so as to improve the generalization ability and robustness of the model.

[0115] Partition of the dataset: The dataset is divided into a training set, a validation set, and a test set, which are used for the dataset to train the model, the dataset for model hyperparameter tuning and avoiding overfitting, and the dataset to finally evaluate the model performance respectively. A custom ratio is used to partition the dataset to build a model that can both learn complex patterns and perform well on unseen data.

[0116] 2. Model architecture

[0117] Model selection: Select YOLOv8 as the base model because of its advantages in real-time performance and accuracy, and improve the model and name it YOLOv8-KK, as Figure 2 shown.

[0118] Network structure: The network structure of YOLOv8 mainly consists of three key parts: the backbone network, the neck network, and the head network. The backbone network is the basis of the model and is responsible for extracting features from the input image. The neck network is located between the backbone network and the head network, and its role is to perform feature fusion and enhancement. The head network is the decision-making part of the object detection model and is responsible for generating the final detection results.

[0119] Backbone network:

[0120] MobileNetV3: Replace the original CSPDarknet53 backbone network with MobileNetV3, which is a lightweight deep neural network. Through hardware-aware neural architecture search (NAS) and NetAdapt algorithm, the performance is further improved. MobileNetV3 combines the inverted residual structure and the Squeeze-and-Excitation (SE) module to achieve efficient feature extraction.

[0121] CA attention mechanism:

[0122] Channel Attention(CA): Introduce the CA attention mechanism based on MobileNetV3. This mechanism enhances the model's perception ability of key features by focusing on the relationships between channels. The CA module is implemented through global average pooling (GAP) and two fully connected layers, which can learn the importance of different channels and re-weight the channel features accordingly to improve the feature expression ability.

[0123] Neck Network:

[0124] Feature Fusion: Path Aggregation Network (PANet) is used as the neck structure, which promotes the information flow between different spatial resolutions, enabling the model to effectively capture multi-scale features.

[0125] Head: The head structure of YOLOv8 consists of multiple detection heads, each of which is responsible for predicting bounding boxes, class probabilities, and objectness scores at different scales.

[0126] Anchor Boxes: Define multi-scale anchor boxes to adapt to object detection of different sizes.

[0127] 3. Loss Function and Optimization

[0128] Loss Function: Propose the EioU loss function as the bounding box regression loss function, which can better handle the overlap and non-overlap situations between target bounding boxes, thereby improving the model's localization accuracy for defects.

[0129] Optimization Algorithm: Adopt optimization algorithms such as Adam or SGD, and adjust the learning rate and other hyperparameters to accelerate the model convergence.

[0130] 4. Training and Validation

[0131] Training Process: Use GPU to accelerate the training process and iteratively train the model with a large amount of data. The training results will obtain a model with the weight containing "Best" (the best-trained model) and "Last" model (the last-trained model).

[0132] Validation and Testing: Evaluate the model performance on independent validation sets and test sets. The evaluation metrics include Accuracy, Precision, Recall, F1 (F1 Score), Confusion Matrix, mAP, etc., to ensure the accuracy and robustness of the model.

[0133] Accuracy: The percentage of correctly predicted results in the total samples. The higher this value, the better the classification effect, and the lower the probabilities of missed detection and misdetection. The calculation format is as shown in Equation 1-1:

[0134]

[0135] TP (True Positive): True positive example, that is, the number of positive samples correctly predicted as positive by the model. For example, the model correctly predicts a cargo that is actually solid waste as solid waste.

[0136] FP (False Positive): False positive example, that is, the number of negative samples incorrectly predicted as positive by the model. For example, the model incorrectly predicts a cargo that is actually not solid waste as solid waste.

[0137] Recall rate: This value represents the probability that the model correctly predicts a positive sample when the actual sample is positive. It reflects the ability of the classifier to detect all true positive samples and is usually used to evaluate the missed detection situation of the model. The calculation formula is as shown in 1-2:

[0138]

[0139] TP (True Positive): True positive example, that is, the number of positive samples correctly predicted as positive by the model. For example, the model correctly predicts a cargo that is actually solid waste as solid waste.

[0140] FN (False Negative): False negative example, that is, the number of positive samples incorrectly predicted as negative by the model. For example, the model incorrectly predicts a cargo that is actually solid waste as non-solid waste.

[0141] The higher the recall rate, the stronger the model's ability to identify positive samples and the lower the probability of missed detection.

[0142] 5. Model Fine-tuning

[0143] Fine-tuning: According to the performance on the validation set, fine-tune the model, optimize the model parameters, and improve the recognition accuracy.

[0144] 6. Integration and Deployment

[0145] System integration: Integrate the trained improved model into the "Intelligent Identification System for Solid Waste Attributes of Electromechanical Products".

[0146] Real-time detection: The model can process the pictures uploaded on-site in real time, score according to the evaluation requirements, and quickly give the identification result.

[0147] 7. Model Evaluation and Iteration

[0148] Performance evaluation: Regularly evaluate the performance of the model in actual applications, including indicators such as accuracy and recall rate.

[0149] Continuous iteration: According to the evaluation results and user feedback, continuously iterate and optimize the model to improve the accuracy and efficiency of identification.

[0150] The functions of the algorithm model are as follows:

[0151] Input: Image data to be detected;

[0152] Output: The detection results of the image, including the scrap rate of the image data, the number of images, the number of various defects, the detection accuracy, the detection recall rate, etc.;

[0153] Declaration name column: Fill in the name of the goods to be identified;

[0154] Supervision method: General trade, inbound repair, import of other items.

[0155] Specifically, it is evaluated from the following several dimensions:

[0156] 1. Usage traces: Brand new and unused: 10 points; Slight usage traces: 5 points; Obvious usage traces: 0 points.

[0157] 2. Packaging form: Individual packaging: 10 points; Simple packaging such as plastic film: 5 points; Mixed packaging or incomplete packaging: 0 points.

[0158] 3. Functional status: Main functions intact: 10 points; Some functions maintained: 5 points; Functions lost: 0 points.

[0159] 4. Appearance inspection: External structure complete and reliable: 10 points; External deformation, damage, surface decoration, cracks, etc.: 5 points; Obvious damage to the external structure, damage to parts: 0 points.

[0160] 5. Safety, health and environmental protection: Meeting safety, health and environmental protection requirements: 10 points; Not meeting: 0 points.

[0161] Combined with on-site photo sampling and uploading to the system and laboratory sampling for detection, a comprehensive score is given to determine the solid waste score line. Above the score line is non-solid waste, and below the score line is solid waste.

[0162] Considering the actual situation, an additional judgment is added at the manual judgment section: Substances not managed as solid waste.

[0163] Step 8:

[0164] Intelligent learning function: When the manual identification conclusion is inconsistent with the system identification conclusion, the manual identification conclusion shall prevail ultimately, and the identification basis shall be input at the same time. When encountering the same type of goods next time, the historical reports will be automatically screened out and used as a reference basis to reduce the inspection risk. Through continuous inspection and accumulation, the accuracy of model identification is improved. When inspecting similar goods, the historical data will be automatically exported through image recognition and other indexes for reference in this inspection, so as to improve the inspection efficiency and reduce errors.

[0165] The intelligent identification method for the solid waste attribute of electromechanical products provided by the present invention uses image recognition technology to replace traditional manual input, improving the information entry efficiency and accuracy; supports the electronicization of original records, improving the on-site inspection efficiency; in addition, the defect detection model automatically determines the uploaded photo information, shields subjective factors to make the identification conclusion more objective, and has an intelligent learning function, collecting data while making a determination, and completing the process of self-learning and self-improvement.

[0166] Based on the same inventive concept, the present invention also provides an intelligent identification system for the solid waste attribute of electromechanical products. Referring to Figure 3 as shown, it includes:

[0167] An OCR recognition module, used to automatically recognize document information through OCR technology, including: goods name, model, quantity, declared value;

[0168] An information collection module, used to obtain photos of goods taken by an image collection device, including: photos of transport packaging, photos of unpacking inspection, photos of label inspection, and photos of key point verification; and obtain the original records during the on-site inspection process;

[0169] A verification module, used to verify the document information with the text information extracted from the photos of the goods taken and the text information of the original records;

[0170] A preliminary judgment module, used when the verification is consistent, according to the determined goods type or scope, to match the document information with the key features extracted from the photos obtained by shooting in a preset database, and make a preliminary judgment and risk assessment of the solid waste attribute of the goods according to the similarity;

[0171] A model detection module, used when the verification is consistent, to input the photos of the goods taken into the defect detection model and output the detection result;

[0172] An identification module, used when the preliminary judgment risk assessment result obtained by the preset data matching is inconsistent with the detection result of the defect detection model, to receive the manual identification result; take the identification result that is consistent between the two or the received manual identification result as the final solid waste attribute identification result.

[0173] This system has a port for uploading inspection information at the port. Before sending personnel to conduct on-site inspection, relevant inspection information such as the stacking situation of goods, transportation situation, unpacking photos, etc. can be uploaded in advance to facilitate the inspection agency to formulate an inspection plan, bring detection tools in advance, and improve the inspection efficiency.

[0174] Taking the solid waste inspection case of a laptop computer shell as an example:

[0175] (1) First, the port entrusting party submits an entrustment application and provides documents, packing list, cargo status description and other information in the system.

[0176] (2) Use OCR technology to automatically identify documents and automatically input relevant field information (all fields on the document) into the system.

[0177] (3) After the inspection agency obtains relevant information from the system and formulates an inspection plan, it will go to the inspection site for inspection, take photos on site, and focus on the inspection of cargo transportation conditions, container information, cargo packaging information, packing information, cargo appearance, and cargo internal structure. Provide information to the system from the following dimensions: stacking form, packaging status, mixed loading form, use traces, structural integrity, functional status, etc. By orally describing the inspection situation, the mobile phone software can be used to record and save it and convert it into text information.

[0178] (4) The relevant inspection photos taken on site are sorted and uploaded to the "Intelligent Identification System for Solid Waste Properties of Electromechanical Products", which makes a comprehensive judgment based on the previous database and identification conclusions. The model calculation is mainly based on the above-mentioned multiple dimensions for scoring. Taking the laptop shell as an example, the focus is on assessing the original function retention, such as whether there are scratches and cracks on the inner and outer surfaces, whether the internal wiring has signs of use, and whether the hinge is intact and corroded. The model scoring principles are as follows: Stacking form: 10 points for neat and orderly, 0 points for random stacking, and 0 points for no packing list. Packaging: 10 points for separate moisture-proof packaging, 0 points for no packaging. Mixed packaging: 10 points for no more than 3 types of goods in the same carton, and 0 points for more than 3 types. 10 points for the inner and outer surfaces intact without cracks, 5 points for scratches, and 0 points for broken or missing. 10 points for the complete function of the wiring hinge, 5 points for signs of use, and 0 points for loss of function. Similar situations are no longer listed, and finally the score is determined to determine whether it is solid waste.

[0179] (5) The inspectors conduct sampling inspections on the laptop computer shells on site based on visual inspection and experience, and after making a comprehensive judgment in accordance with relevant requirements, they form an identification conclusion and enter it into the system. The system compares the two conclusions. When the two conclusions are inconsistent, the manual identification conclusion shall prevail. At the same time, the basis for this judgment and key points of attention shall be supplemented.

[0180] (6) The system completes information collection and updating, and automatically prompts the inspection information of previous years as a reference when uploading the relevant "laptop shell" goods next time. Including time, client information, goods information, import date, inspection date, batch quantity, typical photos, identification conclusion, judgment basis, etc.

[0181] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For related parts, reference can be made to the description in the method section.

[0182] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent identification method for the properties of electromechanical product solid waste, characterized in that: The following steps are involved: S1. Automatically identify document information through OCR technology, including: name, model, quantity, and declared value of goods; S2. Obtain photos of the goods taken by image acquisition equipment, including photos of transport packaging, unpacking inspection, label inspection, and key point inspection; and obtain original records during the on-site inspection process; S3, checking the document information with the text information extracted from the photograph of the goods and the text information of the original record; S4. When the verification is consistent, the document information sheet is matched with the key features extracted from the photographs in a preset database according to the determined type or scope of the goods, and a preliminary judgment and risk assessment of the solid waste properties of the goods are made based on the similarity; S5. When the verification is consistent, the photograph of the goods is input into the defect detection model, and the detection result is output; S6. When the preliminary risk assessment result obtained by matching the preset data is inconsistent with the detection result of the defect detection model, the manual identification result is accepted; the identification result that is consistent between the two or the accepted manual identification result is used as the final solid waste property identification result.

2. The intelligent identification method of electromechanical product solid waste properties according to claim 1 is characterized in that: The step S2 of obtaining the original records during the on-site inspection process includes: Based on the results of the on-site visual inspection, voice description records are made simultaneously, and speech-to-text tools are used to record the audio on-site and automatically generate a text file as the original record.

3. The intelligent identification method of electromechanical product solid waste properties according to claim 1 is characterized in that: The step S3 comprises: Using an exact matching algorithm for some key information, if the information is completely consistent, it is considered a successful match; the key information includes: packing number, model and serial number on the nameplate of the goods, manufacturer, and quantity; The other part of the information is matched using a fuzzy matching algorithm; the other part of the information includes: name of the goods, name of the manufacturer, name of the importer and mode of transportation.

4. The intelligent identification method of electromechanical product solid waste properties according to claim 3 is characterized in that: The step S3 further comprises: When there is a complete mismatch, an alarm will be sent to the inspector and the difference information will be provided; the relevant verification records and original data will be saved.

5. The intelligent identification method of electromechanical product solid waste properties according to claim 1 is characterized in that: The step S4 comprises: When the verification is consistent, the key features extracted from the document and the photograph are matched with the stored data in the preset database using the Pandas match library function algorithm according to the determined type or range of goods; When the matching similarity meets the threshold, the cargo is marked as potential solid waste; and based on the reasons why various products are identified as solid waste extracted from the preset database, a preliminary judgment and risk assessment of the solid waste properties of the cargo is made.

6. The intelligent identification method of electromechanical product solid waste properties according to claim 5 is characterized in that: The preset data in step S4 includes a case library and a knowledge base; Among them, a case library is constructed by collecting solid waste identification cases of electromechanical products over the years; by analyzing the case library, the characteristics of solid waste are refined and a knowledge base is constructed.

7. The intelligent identification method of electromechanical product solid waste properties according to claim 1 is characterized in that: In step S5, the defect detection model is a defect detection and recognition algorithm model improved based on YOLOv8, including a backbone network, a neck network and a head network; Among them, in the backbone network, MobileNetV3 is used to replace the original CSPDarknet53 network; and the CA attention module is introduced based on MobileNetV3; the CA attention module enhances the model's perception of key features by focusing on the relationship between channels.

8. An intelligent identification system for the properties of electromechanical product solid waste, characterized in that: include: OCR recognition module, used to automatically recognize document information through OCR technology, including: cargo name, model, quantity, and declared value; The information collection module is used to obtain photos of the goods taken by image acquisition equipment, including: photos of transport packaging, unpacking inspection photos, identification inspection photos and key point verification photos; and obtain original records during the on-site inspection process; A checking module, used to check the document information sheet with the text information extracted from the photograph of the goods and the text information of the original record; A preliminary judgment module is used to match the document information with the key features extracted from the photographed photos in a preset database when the verification is consistent, according to the determined type or range of the goods, and make a preliminary judgment and risk assessment on the solid waste attributes of the goods based on the similarity; A model detection module, used for inputting the photographed goods into a defect detection model and outputting a detection result when the verification is consistent; The identification module is used to receive manual identification results when the preliminary risk assessment results obtained by matching the preset data are inconsistent with the detection results of the defect detection model; and use the consistent identification results or the received manual identification results as the final solid waste attribute identification results.

Images