Hanging channel machine for streamlined identification of suspended goods in three-dimensional warehouse and identification method

Abstract

The invention relates to the technical field of goods identification, in particular to a hanging channel machine for identifying hanging type goods in a three-dimensional warehouse in a streamlined mode and an identification method, and provides the following scheme that based on a hanging track system, radio frequency labels of the goods are read through a radio frequency scanner to generate a goods list, and appearance images of the goods are collected in combination with a main camera; and determining cargo sorting information and a storage position. A multi-camera system is used for collecting transportation images in real time, sorting information of goods and correctness of track distribution are verified, and if deviation occurs, track distribution is dynamically adjusted, and a conveying path is corrected. Goods storage and distribution are optimized through image segmentation, feature extraction, feature completion and a genetic algorithm, and logistics efficiency is improved. The hanging channel machine is integrated with a multi-sensor system, and efficient identification, sorting, sorting and path adjustment of goods are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of cargo identification, and in particular to a hanging channel machine and an identification method for streamlined identification of hanging cargo in a stereoscopic warehouse. Background Art

[0002] The hanging cargo conveying system realizes the streamlined transportation of cargo through hanging tracks and hanging devices, and has the advantages of high space utilization and adaptability to multiple types of cargo. However, the existing cargo identification and sorting technology still has many limitations in terms of efficiency and accuracy.

[0003] Traditional cargo identification technology usually relies on a single method of barcode or radio frequency tag. Although this method has certain automation advantages in recording cargo information, it has low recognition efficiency for bulk transported goods and cannot correspond one-to-one with the appearance characteristics of the goods, which easily leads to recognition confusion. In addition, the sorting and track allocation of goods still mainly rely on manual experience or fixed rules, which is difficult to dynamically adapt to changes in the actual transportation environment. For example, during transportation, goods may be blocked, stacked or interfered by other uncontrollable factors, resulting in reduced accuracy of recognition and sorting.

[0004] For example, a Chinese patent with authorization announcement number CN111942790B relates to a method for goods entering and exiting a stereoscopic warehouse and exchanging goods between stereoscopic warehouses. The method for entering goods into the warehouse includes: docking the storage space in a transport vehicle with the stereoscopic warehouse, wherein the storage space and the stereoscopic warehouse have one or more storage units; driving a moving device to move an incoming storage device out of a first storage unit in the storage space, and releasing the identity binding relationship between the incoming storage device and the first storage unit, wherein the storage device has goods; and the moving device moves the incoming storage device to a second storage unit in the stereoscopic warehouse, and establishes an identity binding relationship between the incoming storage device and the second storage unit. This invention directly docks two stereoscopic warehouses door to door and layer to layer when goods enter and exit the warehouse, and goods can be transported at different layers at the same time, which is highly efficient and can complete the entry or exit operations of a large number of goods in a short time.

[0005] The above prior art has the problems raised by this background technology: low efficiency and insufficient precision in batch identification, dynamic sorting, real-time verification and path adjustment of hanging cargo, and lack of automatic adjustment capabilities for abnormal transportation processes. To solve the above problems, this application proposes a hanging channel machine and identification method for streamlined identification of hanging cargo in a three-dimensional warehouse. Summary of the invention

[0006] The technical problem to be solved by the present invention is to address the deficiencies of the prior art, and to provide a hanging channel machine and an identification method for streamlined identification of hanging goods in a three-dimensional warehouse. Based on the hanging track system, a radio frequency scanner is used to read the radio frequency tags of the goods to generate a commodity list, and the main camera is used to collect the appearance image of the goods to determine the sorting information and storage location of the goods. A multi-camera system is used to collect transportation images in real time to verify the correctness of the sorting information and track allocation of the goods. If there is a deviation, the track allocation is dynamically adjusted and the transportation path is corrected. The storage and allocation of goods are optimized through image segmentation, feature extraction, feature completion and genetic algorithms to improve logistics efficiency. The channel machine integrates a multi-sensor system to achieve efficient identification, sorting, sorting and path adjustment of goods.

[0007] To achieve the above object, the present invention provides the following technical solutions: The method for identifying suspended cargo in a stereoscopic warehouse in a streamlined manner is applied to a hanging channel machine, wherein the hanging channel machine comprises a hanging track, a hanging device matched with the hanging track, and a plurality of cameras arranged along the hanging track, and a radio frequency scanner and a main camera are arranged at the entrance of the hanging track. The method comprises: Read the commodity radio frequency tags of the goods passing through the hanging track by the radio frequency scanner to generate a commodity list; The main camera is used to collect the appearance image of the hanging cargo at the entrance; Determining goods sorting information according to the goods list and the appearance image, and allocating a storage location for each piece of goods according to the goods sorting information; During the cargo transportation process, multiple cameras arranged along the suspended track are used to collect cargo transportation images, verify the cargo sorting information and the correctness of the track allocation, and if there is a deviation in the cargo sorting, adjust the track allocation and correct the cargo transportation path.

[0008] The determining of the cargo sorting information includes: Separating the cargo area from the background in the appearance image by an image segmentation algorithm, and generating an image matrix containing a plurality of cargo separation areas; performing a sequential segmentation operation on the cargo separation area, arranging the separation area of ​​each cargo in a physical spatial order according to the geometric characteristics of the hanging rail and the relative position of the cargo in the image, and generating a cargo candidate area; Enhance each candidate region, extract regional features of the enhanced candidate region, and calculate a feature set of the cargo candidate region, wherein the regional features include shape features, texture features, and color features; Determine whether the number of the feature set is consistent with the number of goods in the product list. If it is less than the number of goods, supplement the feature set through the feature completion strategy. If it is greater than the number of goods, merge the feature sets through the clustering algorithm. If they are consistent, match the feature set with each product in the product list to generate product sorting information.

[0009] The step of enhancing each candidate region includes: Traverse each pixel in the candidate area, take the pixel as the center point, and the 3×3 pixel matrix around the center point as the window template. Calculate the gray correlation between each pixel in the window template and the center point, and compare the gray correlation with the correlation threshold. If the gray correlation is greater than the correlation threshold, mark the pixel as a correlation point. The number of associated points in the window template is counted. If it is less than four, the center point is marked as a noise point, and the noise point is denoised by grey differential prediction; According to the denoised candidate regions, the shape contours of the candidate regions are extracted layer by layer through a multi-scale pyramid method, and the contour edges are enhanced by combining morphological operations at each scale; Apply wavelet transform to the candidate area after edge enhancement, decompose it into high-frequency texture components and low-frequency texture components, apply texture enhancement algorithm to the high-frequency texture components, apply contrast optimization algorithm to the low-frequency texture components, combine the enhanced high-frequency texture components and low-frequency texture components, and enhance the texture part of the candidate area.

[0010] The feature set for calculating the candidate cargo area includes: By applying directional gradient tensor analysis inside each candidate region, edge curvature, directional distribution and local shape density of the candidate region are extracted, a shape map is constructed according to the analysis results, the shape map is subjected to CT transformation, and shape features are calculated; Extract the centroid of each candidate area, take the centroid as the center, place the first virtual camera at an angle of 30° between the camera and the center plane of the model, take the Z axis as the rotation axis, place the next virtual camera every 30°, and obtain twelve views of the area. The shooting direction of the virtual camera is in the same straight line with the center of the area. Feature extraction is performed on each view. Each view votes for the view feature category according to the view feature classification probability. The view feature category with the most votes is counted as the texture feature of the region. The color information of the candidate area is mapped to the HSV color space and the Lab color space, the histogram information of hue and saturation is extracted in the HSV color space, the brightness and contrast distribution is statistically analyzed in the Lab color space, the color histogram is fused with the block statistical results through a weighted algorithm, and the color features are calculated.

[0011] The feature completion strategy includes: Perform integrity check on the feature set of the cargo candidate area. If the candidate area fails to successfully generate complete shape features, texture features, and color features, the feature of the marked area is in a missing state. The physical position of the missing area is inferred using the spatial distribution information on the suspension track. The initial region prediction box is generated based on the spatial position of the adjacent candidate regions and the camera view angle. The shape, texture, and color information in the initial region prediction box are initialized as blank features. A context-aware network is constructed, and the context features of the initial region prediction box and the missing region are taken as input to generate a feature completion vector for the missing state region.

[0012] The step of allocating a storage location for each piece of cargo includes: According to the goods sorting information, sort the goods list in order of precedence to generate a goods mapping table, wherein the goods mapping table includes goods information and goods order; Collect storage status information, and construct a spatial state matrix S of the three-dimensional storage system based on the storage status information, where each element Indicates the storage space at the coordinate position Status information of the mobile device, wherein the status information includes whether the mobile device is idle, environmental conditions, and neighboring density; Generate a storage demand vector according to the cargo information, traverse the entire space state matrix according to the storage demand vector, calculate an adaptation score for each storage location, and select an initial storage location set according to the adaptation score; Each bin in the initial bin set is iteratively optimized by a genetic algorithm, and a storage position is output.

[0013] The genetic algorithm comprises: Based on each bin in the initial bin set, an initial population is generated, wherein each individual in the population represents a storage location allocation scheme; A fitness function is defined in the genetic algorithm, and the fitness function is used to select the optimal solution according to the frequency of goods entering and leaving the warehouse and the path optimization; The initial solution fitness of each individual is calculated according to the fitness function, and the initial solution fitness is iteratively optimized through selection, crossover and mutation operations until the maximum number of iterations is reached. After each iteration, the initial solution fitness is calculated. If the fitness threshold is met, the solution with the highest fitness is output as the storage location and track path. If the fitness threshold is not met, the fitness function is updated according to the current initial solution fitness.

[0014] Verifying the correctness of the cargo sorting information and track allocation includes: Collecting cargo transportation images according to a plurality of cameras arranged along the suspension track, and extracting cargo position and appearance features of the transportation images through image processing technology; Matching the appearance features with the set of goods features in the sorting information generated at the entrance one by one, calculating the matching confidence, and if the matching confidence is less than the matching threshold, marking the goods sorting information as abnormal; The track position information is obtained according to the cargo position, the path trajectory of the cargo is extracted, the path trajectory is compared with the track path, and it is determined whether the cargo is located on its predetermined track. If not, the path information of the cargo is marked as abnormal.

[0015] The method of adjusting the track allocation and correcting the transport path of the goods includes: According to the cargo marked as abnormal according to the sorting information, the cargo number is recorded, and the cargo mapping table is updated. According to the updated cargo mapping table, a replacement storage location is allocated for the cargo with a location change. According to the replacement storage location and the cargo location, a path adjustment instruction for the cargo is generated, including a target track number and a turning node; According to the goods marked as abnormal in the path information, confirm the current track number and node position of the goods, check whether the target track number of the goods is occupied, and if not, generate a path adjustment instruction; If it is occupied, query the free spare tracks in the track status matrix according to its target storage position, and select the spare track closest to the current track as the replacement track. If there is no spare track, wait for the target track in place and generate a path adjustment instruction.

[0016] A hanging channel machine for streamlining identification of hanging goods in a three-dimensional warehouse, the hanging channel machine comprising: A suspension track system for transporting goods, wherein the suspension track system comprises a suspension track and a suspension device, wherein the suspension track is arranged along a predetermined path and is used to guide the suspended goods to be transported along a set trajectory, and the suspension device cooperates with the suspension track to fix and transport the goods and supports stable suspension and dynamic transportation of the goods; An entrance detection component is used to detect the information of goods to be stored, wherein the entrance detection component includes a radio frequency scanner and a main camera. The radio frequency scanner is configured at the entrance of the hanging track to read the radio frequency tag information of the hanging goods passing through and generate a commodity list containing the unique identification of the goods. The main camera is configured at the entrance of the hanging track to collect the appearance image of the hanging goods and extract the appearance features of the goods. The processing module is configured in the hanging channel machine and is used to execute the above-mentioned streamlined identification method for hanging cargo in a three-dimensional warehouse, including: Sorting information processing: Process the product list generated by the RF scanner and the appearance image collected by the main camera to generate the goods sorting information; Path verification and adjustment: Verify the correctness of cargo sorting information and track allocation based on the transport images collected by multiple cameras arranged along the suspension track, and generate path adjustment instructions when anomalies are detected; Storage location allocation: Based on the goods sorting information and storage status information, the storage location of the goods is optimized using genetic algorithms; Sorting equipment, which is arranged at the end of the hanging track, is used to sort the goods to the corresponding exit or storage location according to the goods sorting information and the path adjustment result; The track status monitoring system collects track status information in real time, including track node occupancy, cargo location and track path information, to support track allocation and path adjustment.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention integrates radio frequency scanners, multi-camera systems and dynamic algorithms, and combines image segmentation, feature extraction, feature completion and other technologies to achieve efficient identification, accurate sorting and dynamic storage location allocation of hanging goods, overcoming the shortcomings of existing technologies in batch identification and sorting accuracy.

[0018] 2. The present invention optimizes storage location allocation by introducing a genetic algorithm, and combines the frequency of goods entering and leaving the warehouse with path optimization to achieve efficient utilization of storage resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings: Figure 1 This is a schematic flow chart of a method for streamlining identification of suspended cargo in a three-dimensional warehouse according to Embodiment 1 of the present invention; Figure 2 This is a flow chart of the clustering algorithm of Example 1 of the present invention; Figure 3 This is a schematic diagram of the texture enhancement process of Example 1 of the present invention; Figure 4 This is a context-aware network structure diagram of Example 1 of the present invention; Figure 5 This is a schematic diagram of the entrance of a hanging channel machine for streamlined identification of hanging cargo in a three-dimensional warehouse according to Example 1 of the present invention. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0021] Embodiment 1: See also Figure 1 The present invention provides an embodiment: a method for streamlining the identification of suspended cargo in a stereoscopic warehouse, wherein the specific steps of the method are as follows: S1: Read the RFID tags of goods through the RFID scanner to generate a list of goods; In this step, the non-contact identification function of the radio frequency scanner is used to quickly obtain the unique identification information of each item. The main purpose of this step is to provide a data basis for subsequent sorting and matching. By generating a list of goods, a unique data record of the goods can be created in the system to achieve full tracking and management of the goods. This method avoids the tediousness and errors of manual registration and improves the speed and accuracy of identification.

[0022] S2: Collect the appearance image of the hanging cargo at the entrance; In this step, by collecting the appearance image of the goods, the visual features of the goods such as shape, color, texture, etc. can be obtained for further sorting and feature matching of goods. RFID tags only provide basic information of goods and cannot be matched one-to-one with goods. Through image processing, further sorting, matching and abnormality detection of goods can be carried out.

[0023] S3: Determine the goods sorting information according to the goods list and the appearance image; In this step, the RFID tag information in the product list is combined with the appearance image of the goods to determine the sorting information of the goods in the current batch through an algorithm. The purpose of this step is to clarify the physical order of the goods before they are transported, so as to provide a basis for subsequent track allocation and storage operations. Through sorting, the logistics operation of batch goods can be transformed into an orderly process to avoid confusion in subsequent operations. At the same time, the sorting information is also a benchmark for checking the transportation track and path of the goods, ensuring that each piece of goods can flow in the correct order.

[0024] S4: Allocate a storage location for each piece of cargo according to the cargo sorting information; In this step, a suitable storage location is assigned to each piece of goods according to the sorting information and storage status of the goods. The main purpose of this step is to optimize the utilization of storage resources while meeting the storage requirements of the goods (such as temperature, humidity, space size, etc.). Through reasonable storage location allocation, the access time of goods can be reduced and the efficiency of storage operations can be improved. In addition, allocating storage locations in combination with sorting information can also ensure that the goods can be quickly matched in subsequent outbound operations, further optimizing the logistics process.

[0025] S5: Collect the transport images of the goods and verify the correctness of the sorting information and track allocation of the goods; In this step, the transport image of the goods is collected by the camera arranged along the hanging track to verify in real time whether the sorting information and track allocation of the goods are correct. The cargo position information and visual features in the transport image can be compared with the sorting information at the entrance to verify whether the goods are transported in the established order. At the same time, the track allocation verification can find out whether the goods have deviated from the predetermined path to ensure the accuracy and stability of the transportation process. This step is a dynamic monitoring of the system operation status, which can detect and handle abnormal situations in time to prevent the problem from spreading further. It should also be noted that due to image occlusion, the initial order of goods may be wrong, resulting in incorrect cargo information. In the case of incorrect information, the matching storage location is likely to be abnormal, such as inability to enter the warehouse or size mismatch, etc. Therefore, it is necessary to verify during transportation to prevent errors in the recognition of the initial image.

[0026] S6: If there is a deviation in the order of the goods, adjust the track allocation and correct the delivery path of the goods; In this step, when deviations in cargo sorting or abnormal paths are detected, the system automatically adjusts track allocation and generates new transport path instructions. The purpose of this step is to ensure that all cargo can be put back into the correct transportation order and eventually reach their target storage location or sorting exit. Through track adjustment, abnormal problems caused by interference or failures during transportation can be dynamically resolved to avoid interruptions to the entire logistics process. At the same time, this step can also optimize the responsiveness of the logistics system, making it more adaptable and flexible to cope with complex transportation environments and diverse cargo needs.

[0027] The specific steps of S3 are as follows: S3.1: Separate the cargo area from the background in the appearance image by an image segmentation algorithm to generate an image matrix containing multiple cargo separation areas; Appearance images may contain complex background information, such as hanging rails, hanging devices, etc. This background information will interfere with the subsequent feature extraction and sorting of goods. Therefore, it is necessary to use image segmentation algorithms to accurately separate the cargo area from the background and retain the visual features related to the cargo; Specifically, the cargo area can be extracted through deep learning-based semantic segmentation algorithms such as U-Net or Mask R-CNN.

[0028] S3.2: performing a sequential segmentation operation on the cargo separation area, arranging the separation area of ​​each cargo in a physical spatial order according to the geometric characteristics of the hanging track and the relative position of the cargo in the image, and generating a cargo candidate area; In the identification of hanging goods, goods are usually arranged in sequence along the hanging track, but due to the limitation of the camera angle or the occlusion of the goods, the arrangement of the goods in the image may be inconsistent with the actual physical order. This will affect the accuracy of the sorting information. Therefore, it is necessary to combine the geometric characteristics of the track (such as track direction, cargo spacing, etc.) and the relative position of the goods in the image to reorder the cargo area and generate candidate areas in the physical space.

[0029] Specifically, according to the design information of the suspension track, the direction vector of the track, the projection relationship between the track and the camera, and the minimum spacing of the cargo on the track are extracted. For each cargo separation area, its centroid coordinates are calculated and projected onto the track direction vector to obtain the projection position of the centroid on the track. According to the size of the projection value, the cargo areas are sorted by projection value from the track entrance to the exit direction to form a physical arrangement of the cargo candidate areas. If the projection values ​​of two areas are too close, it is judged that there may be occlusion, and the area expansion algorithm (such as contour-based expansion operation) is used to merge the overlapping areas.

[0030] S3.3: Enhance each candidate region; In the actual environment, the goods may be affected by uneven lighting, blurred texture, image noise, etc., which will lead to a decrease in the accuracy of feature extraction of the goods. Especially when the appearance features of the goods are similar (such as the same color and material), direct feature extraction may not be able to distinguish different areas of the goods. The purpose of the enhancement operation is to improve the distinguishability of the goods in the image, especially the significance of key features (such as edges and textures), to ensure that the features of the goods can be accurately extracted in complex backgrounds or interference environments.

[0031] S3.4: extracting regional features of the enhanced candidate area and calculating a feature set of the cargo candidate area, wherein the regional features include shape features, texture features and color features; S3.5: Determine whether the number of the feature set is consistent with the number of goods in the product list. If it is less than the number of goods, supplement the feature set through the feature completion strategy. If it is greater than the number of goods, merge the feature sets through the clustering algorithm. If they are consistent, match the feature set with each product in the product list to generate product sorting information.

[0032] Determining whether the number of feature sets is consistent with the product list is a key step to ensure the accuracy of the goods sorting information. If the quantity check is not performed, omissions or redundancies in feature extraction will lead to errors in subsequent sorting or storage allocation. Generating sorting information through feature matching can effectively associate RFID tag information with actual goods.

[0033] Specifically, the feature set is clustered using the DBSCAN and K-means clustering algorithms, see Figure 2 , the flow chart of the clustering algorithm of the embodiment of the present invention, firstly, the number of core points is set according to the number of commodities, the features in the feature set are preliminarily clustered by the DBSCAN clustering algorithm, the core points are divided according to the preliminary clustering results, the core points are used as the initial center clusters of the K-means clustering algorithm, the Euclidean distance from the feature set to each initial center cluster is calculated, the clustering is divided according to the Euclidean distance, and the average clustering error of the initial center cluster is calculated, the center cluster is updated according to the average clustering error, the convergence condition is defined, and it is judged whether the updated center cluster meets the convergence condition. If the convergence condition is not met, the center cluster is continuously updated. If the convergence condition is met, the clustering result is output.

[0034] S3.3 The specific steps are as follows: S3.3.1: Traverse each pixel in the candidate area, take the pixel as the center point, and the 3×3 pixel matrix around the center point as the window template, calculate the gray correlation between each pixel in the window template and the center point, and compare the gray correlation with the correlation threshold. If the gray correlation is greater than the correlation threshold, mark the pixel as a correlation point; The image in the candidate area may contain background noise, blurred edges and other factors, which will interfere with the accuracy of subsequent feature extraction. By traversing each pixel and analyzing the gray correlation of its surrounding pixels, the important pixels and noise pixels in the target area can be effectively distinguished. Gray correlation is an indicator that reflects the similarity of pixel values ​​and can evaluate the similarity between pixels in a local range.

[0035] S3.3.2: Count the number of associated points in the window template. If it is less than four, mark the center point as a noise point and denoise the noise point by grey differential prediction; When the number of associated points is small, the center point is likely to be noise or isolated pixels, and retaining these pixels will interfere with the subsequent shape contour extraction. Processing noise points can further improve image quality and feature extraction accuracy.

[0036] S3.3.3: Based on the denoised candidate regions, the shape contours of the candidate regions are extracted layer by layer using a multi-scale pyramid method, and the contour edges are enhanced by combining morphological operations at each scale; By extracting shape contours layer by layer through the multi-scale pyramid method, regional features can be analyzed from different resolution levels, so that small-scale details and large-scale structures are retained at the same time. At the same time, by combining morphological operations to enhance the edges, discontinuous contour breakpoints and locally weakened edge features can be eliminated, thereby ensuring that the extracted contours are complete and clear.

[0037] Specifically, by Gaussian blurring and downsampling the candidate area images, multi-scale pyramid images are generated layer by layer. Each layer of the image retains regional information of different resolutions, captures the overall contour of a large range in the low-resolution layer, and captures the edge information of details in the high-resolution layer. In each layer of the pyramid image, an edge detection algorithm is used to extract preliminary contours, and morphological operations are performed on the extracted contours, including dilation and erosion, to strengthen the connections at the breakpoints and remove pseudo edges. The contour information extracted from different scales is merged layer by layer, and the optimal features of multi-scale information are retained through the edge strength weighted fusion method. For the merged contours, morphological closing operations are further applied to eliminate small gaps in the contours to ensure the integrity of the contours.

[0038] S3.3.4: Apply wavelet transform to the candidate region after edge enhancement to decompose it into high-frequency texture components and low-frequency texture components, apply texture enhancement algorithm to the high-frequency texture components, apply contrast optimization algorithm to the low-frequency texture components, combine the enhanced high-frequency texture components and low-frequency texture components, and enhance the texture part of the candidate region; Texture features are an important part of the appearance of goods and directly affect the accuracy of feature matching. However, since images may be affected by light, noise, etc., high-frequency textures may not be significant enough and low-frequency textures may appear to be insufficiently contrasted. By decomposing textures into high-frequency and low-frequency components through wavelet transform, the details and overall performance of texture features can be optimized respectively, ultimately achieving comprehensive enhancement of texture information.

[0039] See also Figure 3, a schematic diagram of the texture enhancement process of an embodiment of the present invention, firstly, wavelet transform is applied to the candidate region after edge enhancement, and the image is decomposed into high-frequency texture components (representing detail information, such as edges and fine textures) and low-frequency texture components (representing overall structure and uniform texture). The decomposed components are stored in the form of image subbands, representing the texture characteristics in the horizontal, vertical and diagonal directions respectively. The texture enhancement algorithm is applied to the high-frequency texture component, and the significance of the edge is enhanced by nonlinear stretching operation, while suppressing the influence of high-frequency noise. The adaptive filter is used in the detail area to further enhance the fine texture features, making it more prominent in the complex background. The contrast optimization algorithm is applied to the low-frequency texture component to adjust the brightness and contrast distribution to improve the distinguishability of the uniform texture, and the smoothing filter is used to eliminate the pseudo texture and uneven illumination in the low-frequency component. The enhanced high-frequency texture component and the low-frequency texture component are subjected to wavelet inverse transform to generate the enhanced texture image of the candidate region. The high-frequency texture enhancement highlights the detailed features of the cargo surface, which helps to distinguish similar objects, while the low-frequency texture optimization improves the overall quality of the texture distribution, making the texture characteristics of the candidate region clearer and more consistent.

[0040] S3.4 The specific steps are as follows: S3.4.1: extract edge curvature, orientation distribution, and local shape density of the candidate region by applying directional gradient tensor analysis inside each candidate region, construct a shape map based on the analysis results, perform CT transformation on the shape map, and calculate shape features; The main purpose of directional gradient tensor analysis is to extract the geometric features of the candidate area, including edge curvature, directional distribution, and local shape density. These features can better describe the shape contour and spatial characteristics of the goods and provide accurate input for subsequent shape feature extraction. Shape features are crucial for distinguishing different goods because they can reflect the appearance structure and spatial characteristics of the goods and are directly related to shape matching in the visual recognition system. By constructing a shape graph and performing transformation analysis, complex shape features can be uniformly represented, improving the system's ability to recognize diverse goods.

[0041] Specifically, within the candidate region, the gradient direction and intensity of each pixel are calculated by the gradient operator to characterize the direction and change amplitude of the local edge. Gradient tensor analysis captures the overall characteristics of the directional distribution by integrating the gradient information of the pixel into a tensor matrix. For example, for the gradient of a certain region, the tensor can quantify the main direction of the gradient and its intensity distribution in the region. During the analysis, the edge curvature is used to characterize the curvature of the edge, which helps to identify the complex contour features of the goods; the directional distribution is used to quantify the main shape direction of the candidate region; the local shape density indicates the distribution density of the edge within a certain range, which helps to determine whether the shape is regular or blocked. Combining the edge curvature, directional distribution and local shape density, a two-dimensional shape map is generated. The value of each pixel in the shape map indicates the importance of the point in the shape feature. For example, the edge region with high curvature appears as a significantly high value in the shape map, which enables the system to give priority to the region with important shape information. CT transformation is an analysis method that transforms the shape map from the spatial domain to the feature domain. Through CT transformation, the potential global and local characteristics in the shape graph can be extracted. By decomposing the local geometric characteristics in the shape graph, it is converted into a feature vector. Each component of the feature vector reflects the response intensity of the shape graph in a certain feature dimension.

[0042] S3.4.2: Extract the centroid of each candidate area, place the first virtual camera at an angle of 30° between the camera and the center plane of the model with the centroid as the center, and place the next virtual camera every 30° with the Z axis as the rotation axis to obtain twelve views of the area, where the shooting direction of the virtual camera is in the same straight line as the center of the area; S3.4.3: Extract features for each view, and each view votes for the view feature category according to the view feature classification probability, and the view feature category with the most votes is counted as the texture feature of the region; Specifically, through multi-view feature extraction and classification, the feature performance of the candidate region under different viewing angles can be comprehensively utilized to effectively improve the robustness and accuracy of classification. The voting mechanism can smooth the randomness of single-view classification and ensure that the final texture feature category can truly reflect the overall characteristics of the candidate region. In addition, this multi-view classification method has significant adaptability and anti-interference ability for candidate regions with complex shapes or occlusions.

[0043] S3.4.4: Map the color information of the candidate area to the HSV color space and the Lab color space, extract the histogram information of hue and saturation in the HSV color space, perform block statistics on the brightness and contrast distribution in the Lab color space, fuse the color histogram with the block statistics results through a weighted algorithm, and calculate the color features.

[0044] Color features are a very important dimension in visual recognition, which can help distinguish goods with different color distributions. In practical applications, directly using the original RGB color information may cause color distortion due to changes in lighting conditions, affecting the stability of feature extraction. By mapping color information to HSV and Lab color spaces, the effects of hue, saturation and brightness can be better separated, and the robustness to lighting changes can be enhanced. In addition, combining features from different color spaces helps capture more color details and improve the accuracy of classification and matching.

[0045] If a complete feature set is not generated, the goods cannot be accurately identified and managed, so it is necessary to complete it. The specific steps of the feature completion strategy are as follows: S3.5.1: Perform integrity check on the feature set of the cargo candidate area. If the candidate area fails to successfully generate complete shape features, texture features, and color features, the marked area features are in a missing state; Use the threshold detection algorithm to check the continuity of the shape contour. If there is a break or blur, it is marked as missing. The texture intensity is detected through the frequency domain distribution of the texture feature. If the texture signal is too weak or irregular, it is marked as missing. For color features, check whether there is abnormal deviation through the color histogram. If it deviates from the normal distribution, it is marked as missing, identifying feature extraction failures caused by occlusion, insufficient lighting, or motion blur.

[0046] S3.5.2: Use the spatial distribution information on the suspension track to infer the physical position of the missing area, and use the spatial position of the adjacent candidate areas and the camera view as constraints to generate an initial region prediction box, and initialize the shape, texture, and color information within the initial region prediction box as blank features.

[0047] In this step, the possible location of the missing area is estimated based on the cargo spacing and transportation sequence information on the track; secondly, the size and shape of the prediction box are calculated using the perspective relationship of the camera view; finally, the features in the prediction box are initialized as blank features so that they can be filled in later through the completion algorithm, which constrains the spatial range of the missing area and improves the efficiency and accuracy of completion.

[0048] Specifically, the geometric characteristics of the suspension track and the camera's shooting angle are used to infer the physical location of the area where the current feature is missing. By analyzing the positional relationship of adjacent candidate areas on the track, the possible spatial distribution range is determined. For example, if the goods in the front and rear areas are evenly distributed on the track, the missing area may be located in a certain section in the middle of the track. Combined with the camera's angle of view and resolution, a preliminary prediction box is generated to cover the possible missing area.

[0049] S3.5.3: Construct a context-aware network, take the initial region prediction box and the context features of the missing region as input, and generate a feature completion vector for the missing state region.

[0050] See also Figure 4 , a context-aware network structure diagram of an embodiment of the present invention, wherein the context-aware network is used to process a deep learning model for completing features of candidate cargo regions, and the network structure is provided with an input layer, a context extraction layer, and an output layer, specifically including: The input layer is used to receive the context features of the initial region prediction box and the adjacent regions, including shape features, texture features, and color features. First, these features are preprocessed to standardize the distribution of the input data and reduce the feature dimension to adapt to the input requirements of the context-aware network. The processed feature matrix is ​​used as the initial input of the network to provide data support for subsequent feature completion operations.

[0051] The context extraction layer extracts the depth information of the initial region prediction box and context features through the convolutional network and attention mechanism, and generates the feature descriptors required for completion.

[0052] First, the context feature matrix is ​​scanned using multi-scale convolution kernels to extract local shape features, texture features, and color features. The convolution operation processes the feature matrix through multiple convolution kernels of different sizes to capture multi-scale context information. These features are combined into a deep feature map to represent the spatial and visual correlation features of the missing area. The extracted deep feature map is weighted according to the channel attention mechanism and spatial attention mechanism.

[0053] The channel attention mechanism analyzes the importance of each feature channel, increases the weight of feature channels that are highly relevant to the missing area, and suppresses the weight of irrelevant or low-relevant channels.

[0054] The spatial attention mechanism assigns spatial weights according to the contextual relevance of each position in the deep feature map, enhancing the position features with high spatial correlation with the missing area.

[0055] The deep feature map and attention-optimized features are fused and a comprehensive contextual feature descriptor is generated through a weighted concatenation operation to fully describe the shape, texture, and color features of the missing area.

[0056] The output layer is used to process the feature descriptors generated by the context extraction layer and generate the final feature completion results.

[0057] First, the context feature descriptor is fully connected to map the multidimensional feature vector to the component space of shape features, texture features, and color features. The mapping process ensures that the completed features are consistent with the requirements of the missing area and generates the missing features, including: The shape contour of the missing area is generated through the deep learning decoding module to complete the contour break problem caused by the missing area. According to the texture direction information of the context, the completed texture is generated that is consistent with the texture of the surrounding area. Through the color distribution prediction algorithm, the completed color features that are consistent with the color distribution of the context area are generated.

[0058] The generated shape, texture, and color features are weighted and concatenated to form a complete feature completion vector. To ensure the reliability of the completion result, the output layer uses the RELU activation function to nonlinearly map the feature values ​​and uses the sigmoid activation function to calculate the confidence of the completion feature. If the confidence is lower than the preset threshold, the output layer will readjust the weight of the context feature for completion optimization; if the confidence reaches or exceeds the threshold, the final feature completion vector is output.

[0059] The specific steps of S4 are as follows: S4.1: sorting the commodity list in order according to the commodity sorting information to generate a commodity mapping table, wherein the commodity mapping table includes commodity information and commodity order; The cargo mapping table is the basic data structure for the subsequent storage location allocation. It associates the sorting information of each cargo with its characteristic information to ensure that the storage allocation process has a clear input basis. Through sorting, storage and warehousing operations can be optimized, making the management of cargo more efficient and systematic.

[0060] Specifically, according to the cargo sorting information, each cargo is traversed in the order of the commodity list, and the RFID tag data, appearance characteristics and generated sorting position information of each cargo are integrated into a data record. Then these data records are stored in a table structure (cargo mapping table) in the sorting order. The table contains the cargo number, sorting information, storage condition requirements (such as temperature, humidity, volume, etc.), and is combined with the subsequent storage space status to provide input for the storage allocation algorithm.

[0061] S4.2: Collect storage status information, and construct a spatial state matrix S of the three-dimensional storage system based on the storage status information, where each element Indicates the storage space at the coordinate position Status information of the mobile device, wherein the status information includes whether the mobile device is idle, environmental conditions, and neighboring density; The status information of storage space directly determines the feasibility of cargo storage allocation. By constructing a space status matrix, the usage of storage locations, environmental adaptability, and proximity density information can be grasped in real time, thereby providing comprehensive status support for storage allocation calculations. Through real-time updated matrices, the system can quickly screen out qualified storage locations, improve allocation efficiency, and reduce storage space waste.

[0062] S4.3: Generate a storage demand vector according to the cargo information, traverse the entire space state matrix according to the storage demand vector, calculate the adaptation score for each storage location, and select the initial storage location set according to the adaptation score, wherein the calculation formula of the adaptation score is: , Among them, A represents the adaptation score of the storage location, Q represents the state value of the storage location. If the storage location has stored goods, Q=0, otherwise Q=1. W represents the weight matching value between the storage location carrying capacity and the goods. If the weight of the goods is greater than the carrying capacity, W=0, otherwise W=1. , and represents the weighting coefficient, Indicates the size of the goods. represents the storage location size, q represents the sensitivity coefficient of environmental conditions matching, Indicates the storage conditions for goods. Indicates the environmental conditions of the storage location, Represents a condition matching function, represents the expected proximity density of the cargo, represents the storage location proximity density, The smoothing factor representing the neighboring density is used to control the decay rate of the density adaptation; The initial storage location set is the preliminary screening result of optimizing the storage location allocation. Its purpose is to quickly narrow the search scope in the entire storage space and improve the efficiency of storage location allocation. According to the storage condition requirements of each piece of goods in the cargo mapping table, the corresponding storage demand vector is generated, including size, load weight, temperature and humidity requirements. Combined with the storage demand vector, the spatial state matrix is ​​traversed to calculate the degree of adaptation of each storage location one by one. In the process of calculating the adaptation score, the idle state of the storage location, the degree of matching between the environmental conditions and the cargo requirements, and whether the neighboring density meets the protection requirements of the cargo are comprehensively considered. The storage locations with higher adaptation scores are added to the initial storage location set to form a list of candidate storage locations. It should be noted that the initial screening focuses on the matching of local conditions, such as whether the storage location is idle and whether the size matches, but it cannot optimize the resource utilization of the entire storage system.

[0063] S4.4: Iteratively optimize each storage location in the initial storage location set by using a genetic algorithm, and output a storage location; The initial selection of storage locations is based on a single fitness score, without in-depth analysis of comprehensive factors such as the global state of storage, cargo priority, and track path efficiency. Therefore, the initial selection of storage locations may contain multiple storage locations, which, although meeting basic needs, may not be the best storage options. Through iterative optimization of genetic algorithms, the optimal storage location can be further found in the candidate storage locations to meet the multi-objective requirements of cargo distribution (such as storage efficiency, storage balance, etc.).

[0064] Specifically, the candidate locations in the initial location set are used as the search space, and several storage location allocation schemes are randomly generated as the initial population, each of which represents the correspondence between goods and locations. The fitness function is defined in the genetic algorithm, and a comprehensive score is given based on the frequency of goods entering and leaving the warehouse and the efficiency of the track path. Then, the population is iteratively optimized through operations such as selection, crossover, and mutation. In each iteration, the scheme with higher fitness is retained, and the scheme with lower fitness is eliminated until the optimization goal is met or the set number of iterations is reached. Finally, the scheme with the highest fitness is output as the storage allocation result.

[0065] The genetic algorithm comprises: Based on each bin in the initial bin set, an initial population is generated, wherein each individual in the population represents a storage location allocation scheme; A fitness function is defined in the genetic algorithm, and the fitness function is used to select the optimal solution according to the frequency of goods entering and leaving the warehouse and the path optimization; The initial solution fitness of each individual is calculated according to the fitness function, and the initial solution fitness is iteratively optimized through selection, crossover and mutation operations until the maximum number of iterations is reached. After each iteration, the initial solution fitness is calculated. If the fitness threshold is met, the solution with the highest fitness is output as the storage location and track path. If the fitness threshold is not met, the fitness function is updated according to the current initial solution fitness.

[0066] The specific steps of S5 are as follows: S5.1: collecting a transport image of the cargo according to a plurality of cameras arranged along the suspension track, and extracting the cargo position and appearance features of the transport image by image processing technology; Specifically, the location of the cargo can be identified through deep learning models, such as YOLOv5, to identify the area where the cargo is located and generate a bounding box for each piece of cargo to ensure that the cargo location can be accurately identified in complex transportation scenarios. The appearance features of the cargo are then extracted within the bounding box, including color histograms, texture patterns (such as local binary patterns), and shape contour features, to form the appearance features of the cargo.

[0067] S5.2: Match the appearance features with the set of goods features in the sorting information generated at the entrance one by one, calculate the matching confidence, and if the matching confidence is less than the matching threshold, mark the goods sorting information as abnormal; Sorting information directly affects the track allocation, route planning and storage location allocation of goods. If the sorting information is wrong, the goods will be allocated to the wrong storage location, which may result in the goods not being properly stored, storage resources being wasted, or even the logistics process being interrupted. Rechecking the sorting information is to verify whether the sorting of goods during transportation is correct and to ensure the efficiency and accuracy of the entire logistics process. Usually, abnormal sorting information may be caused by the following reasons: The entrance image taken by the main camera is affected by occlusion or light changes, resulting in incomplete feature extraction; due to the high similarity of appearance features between batches of goods, mismatching may occur, leading to sorting errors; the record of the goods feature list in the previous link may be deviated due to image segmentation errors or calculation errors.

[0068] Specifically, incorrect sorting information will cause the warehousing system to allocate and schedule based on erroneous data. For example, a piece of goods will be delivered to a storage location that is not suitable for its storage conditions, thus affecting the entire logistics process. Therefore, by collecting real-time images through cameras during transportation and re-detecting sorting information, defects in the initial sorting can be compensated, ensuring that the goods always maintain the correct logical order in a dynamic transportation environment.

[0069] S5.3: Acquire the track position information according to the cargo position, extract the path track of the cargo, compare the path track with the track path, and determine whether the cargo is on its predetermined track. If not, mark the path information of the cargo as abnormal.

[0070] The specific steps of S6 are as follows: S6.1: Record the cargo number of the cargo marked as abnormal according to the sorting information, and update the cargo mapping table. According to the updated cargo mapping table, assign a replacement storage location to the cargo with a location change. According to the replacement storage location and the cargo location, generate a path adjustment instruction for the cargo, including a target track number and a turning node. S6.2: for the goods marked as abnormal according to the path information, confirm the track number and node position where the goods are currently located, check whether the target track number of the goods is occupied, and if not, generate a path adjustment instruction; S6.3: If it is occupied, query the free spare track in the track status matrix according to its target storage position, and select the spare track closest to the current track as the replacement track. If there is no spare track, wait for the target track in place and generate a path adjustment instruction.

[0071] Embodiment 2: See also Figure 5 , the present invention provides an embodiment: a hanging channel machine for streamlining identification of hanging goods in a three-dimensional warehouse, the figure shows the three-dimensional channel structure on both sides, the hanging track runs through the channel to form a space guide area, the dotted arrows indicate the movement path of the goods along the track, gradually transported from the entrance to the exit position or the target storage position, reflecting the flow direction and preset trajectory of the goods; The mounting channel machine includes: A suspension track system for transporting goods, wherein the suspension track system comprises a suspension track and a suspension device, wherein the suspension track is arranged along a predetermined path and is used to guide the suspended goods to be transported along a set trajectory, and the suspension device cooperates with the suspension track to fix and transport the goods and supports stable suspension and dynamic transportation of the goods; An entrance detection component is used to detect the information of goods to be stored, wherein the entrance detection component includes a radio frequency scanner and a main camera. The radio frequency scanner is configured at the entrance of the hanging track to read the radio frequency tag information of the hanging goods passing through and generate a commodity list containing the unique identification of the goods. The main camera is configured at the entrance of the hanging track to collect the appearance image of the hanging goods and extract the appearance features of the goods. The processing module is configured in the mounting channel machine and includes: Sorting information processing: Process the product list generated by the RF scanner and the appearance image collected by the main camera to generate the goods sorting information; Path verification and adjustment: Verify the correctness of cargo sorting information and track allocation based on the transport images collected by multiple cameras arranged along the suspension track, and generate path adjustment instructions when anomalies are detected; Storage location allocation: Based on the goods sorting information and storage status information, the storage location of the goods is optimized using genetic algorithms; Sorting equipment, which is arranged at the end of the hanging track, is used to sort the goods to the corresponding exit or storage location according to the goods sorting information and the path adjustment result; The track status monitoring system collects track status information in real time, including track node occupancy, cargo location and track path information, to support track allocation and path adjustment.

[0072] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for streamlining the identification of suspended cargo in a three-dimensional warehouse, which is applied to a hanging channel machine, wherein the hanging channel machine comprises a hanging track, a hanging device matched with the hanging track, and a plurality of cameras arranged along the hanging track, wherein a radio frequency scanner and a main camera are arranged at the entrance of the hanging track, and the characteristics are as follows: The method comprises: Read the commodity radio frequency tags of the goods passing through the hanging track by the radio frequency scanner to generate a commodity list; The main camera is used to collect the appearance image of the hanging cargo at the entrance; Determine the goods sorting information according to the goods list and the appearance image, and assign a storage location to each piece of goods according to the goods sorting information, wherein determining the goods sorting information includes: separating the goods area from the background by an image segmentation algorithm, obtaining a goods candidate area, performing feature extraction on the goods candidate area, generating a feature set, determining whether the number of the feature set is consistent with the number of goods in the goods list, if it is less than the number of goods, supplementing the feature set by a feature completion strategy, if it is greater than the number of goods, merging the feature sets by a clustering algorithm, and if they are consistent, matching the feature set with each of the goods in the goods list to generate the goods sorting information; During the cargo transportation process, multiple cameras arranged along the suspended track are used to collect cargo transportation images, verify the cargo sorting information and the correctness of the track allocation, and if there is a deviation in the cargo sorting, adjust the track allocation and correct the cargo transportation path.

2. The method for identifying suspended cargo in a three-dimensional warehouse in a streamlined manner according to claim 1 is characterized in that: The determining of the cargo sorting information includes: Separating the cargo area from the background in the appearance image by an image segmentation algorithm, and generating an image matrix containing a plurality of cargo separation areas; performing a sequential segmentation operation on the cargo separation area, arranging the separation area of ​​each cargo in a physical spatial order according to the geometric characteristics of the hanging rail and the relative position of the cargo in the image, and generating a cargo candidate area; Enhance each candidate region, extract regional features of the enhanced candidate region, and calculate a feature set of the cargo candidate region, wherein the regional features include shape features, texture features, and color features; Determine whether the number of the feature set is consistent with the number of goods in the product list. If it is less than the number of goods, supplement the feature set through the feature completion strategy until it is consistent with the number of goods in the product list. If it is greater than the number of goods, merge the feature sets through the clustering algorithm until it is consistent with the number of goods in the product list. If they are consistent, the feature set is matched with each product in the product list to generate product sorting information.

3. The method for streamlining the identification of suspended cargo in a three-dimensional warehouse according to claim 2 is characterized in that: The step of enhancing each candidate region includes: Traverse each pixel in the candidate area, take the pixel as the center point, and the 3×3 pixel matrix around the center point as the window template. Calculate the gray correlation between each pixel in the window template and the center point, and compare the gray correlation with the correlation threshold. If the gray correlation is greater than the correlation threshold, mark the pixel as a correlation point. The number of associated points in the window template is counted. If it is less than four, the center point is marked as a noise point, and the noise point is denoised by grey differential prediction; According to the denoised candidate regions, the shape contours of the candidate regions are extracted layer by layer through a multi-scale pyramid method, and the contour edges are enhanced by combining morphological operations at each scale; Apply wavelet transform to the candidate area after edge enhancement, decompose it into high-frequency texture components and low-frequency texture components, apply texture enhancement algorithm to the high-frequency texture components, apply contrast optimization algorithm to the low-frequency texture components, combine the enhanced high-frequency texture components and low-frequency texture components, and enhance the texture part of the candidate area.

4. The method for identifying suspended cargo in a three-dimensional warehouse in a streamlined manner according to claim 3 is characterized in that: The feature set for calculating the candidate cargo area includes: By applying directional gradient tensor analysis inside each candidate region, edge curvature, directional distribution and local shape density of the candidate region are extracted, a shape map is constructed according to the analysis results, the shape map is subjected to CT transformation, and shape features are calculated; Extract the centroid of each candidate area, take the centroid as the center, place the first virtual camera at an angle of 30° between the camera and the center plane of the model, take the Z axis as the rotation axis, place the next virtual camera every 30°, and obtain twelve views of the area. The shooting direction of the virtual camera is in the same straight line with the center of the area. Feature extraction is performed on each view. Each view votes for the view feature category according to the view feature classification probability. The view feature category with the most votes is counted as the texture feature of the region. The color information of the candidate area is mapped to the HSV color space and the Lab color space, the histogram information of hue and saturation is extracted in the HSV color space, the brightness and contrast distribution is statistically analyzed in the Lab color space, the color histogram is fused with the block statistical results through a weighted algorithm, and the color features are calculated.

5. The method for identifying suspended cargo in a three-dimensional warehouse in a streamlined manner according to claim 2 is characterized in that: The feature completion strategy includes: Perform integrity check on the feature set of the cargo candidate area. If the candidate area fails to successfully generate complete shape features, texture features, and color features, the feature of the marked area is in a missing state. The physical position of the missing area is inferred using the spatial distribution information on the suspension track. The initial region prediction box is generated based on the spatial position of the adjacent candidate regions and the camera view angle. The shape, texture, and color information in the initial region prediction box are initialized as blank features. A context-aware network is constructed, and the context features of the initial region prediction box and the missing region are taken as input to generate a feature completion vector for the missing state region.

6. The method for identifying suspended cargo in a three-dimensional warehouse according to claim 1 is characterized in that: The step of allocating a storage location for each piece of cargo includes: According to the goods sorting information, sort the goods list in order of precedence to generate a goods mapping table, wherein the goods mapping table includes goods information and goods order; Collect storage status information, and construct a spatial state matrix S of the three-dimensional storage system based on the storage status information, where each element Indicates the storage space at the coordinate position Status information of the mobile device, wherein the status information includes whether the mobile device is idle, environmental conditions, and neighboring density; Generate a storage demand vector according to the cargo information, traverse the entire space state matrix according to the storage demand vector, calculate an adaptation score for each storage location, and select an initial storage location set according to the adaptation score; Each bin in the initial bin set is iteratively optimized by a genetic algorithm, and a storage position is output.

7. The method for streamlining the identification of suspended cargo in a three-dimensional warehouse according to claim 6 is characterized in that: The genetic algorithm comprises: Based on each bin in the initial bin set, an initial population is generated, wherein each individual in the population represents a storage location allocation scheme; A fitness function is defined in the genetic algorithm, and the fitness function is used to select the optimal solution according to the frequency of goods entering and leaving the warehouse and the path optimization; The initial solution fitness of each individual is calculated according to the fitness function, and the initial solution fitness is iteratively optimized through selection, crossover and mutation operations until the maximum number of iterations is reached. After each iteration, the initial solution fitness is calculated. If the fitness threshold is met, the solution with the highest fitness is output as the storage location and track path. If the fitness threshold is not met, the fitness function is updated according to the current initial solution fitness.

8. The method for streamlining the identification of suspended cargo in a three-dimensional warehouse according to claim 7 is characterized in that: Verifying the correctness of the cargo sorting information and track allocation includes: Collecting cargo transportation images according to a plurality of cameras arranged along the suspension track, and extracting cargo position and appearance features of the transportation images through image processing technology; Matching the appearance features with the set of goods features in the sorting information generated at the entrance one by one, calculating the matching confidence, and if the matching confidence is less than the matching threshold, marking the goods sorting information as abnormal; The track position information is obtained according to the cargo position, the path trajectory of the cargo is extracted, the path trajectory is compared with the track path, and it is determined whether the cargo is located on its predetermined track. If not, the path information of the cargo is marked as abnormal.

9. The method for streamlining the identification of suspended cargo in a three-dimensional warehouse according to claim 8 is characterized in that: The method of adjusting the track allocation and correcting the transport path of the goods includes: According to the cargo marked as abnormal according to the sorting information, the cargo number is recorded, and the cargo mapping table is updated. According to the updated cargo mapping table, a replacement storage location is allocated for the cargo with a location change. According to the replacement storage location and the cargo location, a path adjustment instruction for the cargo is generated, including a target track number and a turning node; According to the goods marked as abnormal in the path information, confirm the current track number and node position of the goods, check whether the target track number of the goods is occupied, and if not, generate a path adjustment instruction; If it is occupied, query the free spare tracks in the track status matrix according to its target storage position, and select the spare track closest to the current track as the replacement track. If there is no spare track, wait for the target track in place and generate a path adjustment instruction.

10. The hanging channel machine for streamlined identification of hanging goods in three-dimensional warehouses is characterized by: The hanging channel machine comprises: A suspension track system for transporting goods, wherein the suspension track system comprises a suspension track and a suspension device, wherein the suspension track is arranged along a predetermined path and is used to guide the suspended goods to be transported along a set trajectory, and the suspension device cooperates with the suspension track to fix and transport the goods and supports stable suspension and dynamic transportation of the goods; An entrance detection component is used to detect the information of goods to be stored, wherein the entrance detection component includes a radio frequency scanner and a main camera. The radio frequency scanner is configured at the entrance of the hanging track to read the radio frequency tag information of the hanging goods passing through and generate a commodity list containing the unique identification of the goods. The main camera is configured at the entrance of the hanging track to collect the appearance image of the hanging goods and extract the appearance features of the goods. The processing module is configured in the hanging channel machine, and is used to execute the method for identifying the hanging cargo in the stereoscopic warehouse in a streamlined manner as claimed in any one of claims 1 to 9, comprising: Sorting information processing: Process the product list generated by the RF scanner and the appearance image collected by the main camera to generate the goods sorting information; Path verification and adjustment: Verify the correctness of cargo sorting information and track allocation based on the transport images collected by multiple cameras arranged along the suspension track, and generate path adjustment instructions when anomalies are detected; Storage location allocation: Based on the goods sorting information and storage status information, the storage location of the goods is optimized using genetic algorithms; Sorting equipment, which is arranged at the end of the hanging track, is used to sort the goods to the corresponding exit or storage location according to the goods sorting information and the path adjustment result; The track status monitoring system collects track status information in real time, including track node occupancy, cargo location and track path information, to support track allocation and path adjustment.

Images