Intelligent warehouse-out management method and device

Abstract

The invention discloses a warehouse intelligent ex-warehouse management method and device, and relates to the technical field of warehouse management, and the method comprises the steps: determining a target type of ex-warehouse goods according to ex-warehouse order information; images of all goods corresponding to the target type in the warehouse are obtained, and target goods are screened out; the method comprises the following steps: calculating a picking difficulty value of target goods, establishing a two-dimensional coordinate system according to a warehouse, obtaining a coordinate corresponding to each target goods, carrying out route planning in combination with a warehouse entrance coordinate and the picking difficulty value of the target goods, obtaining an optimal picking path, and sending the optimal picking path to a picker. According to the goods picking method, workers can pick goods according to the optimal goods picking path, and do not need to frequently shuttle among different goods shelves when picking the goods, so that the order of the goods picking process is ensured, the fatigue of the workers is reduced, the accuracy rate of goods picking is ensured, and the conditions of mistakenly picking goods, missing goods or placing the goods in wrong positions are reduced; and the negative influence on the warehouse-out efficiency and the overall operation of the warehouse is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of warehouse management, and in particular to a method and device for intelligent warehouse delivery management. Background Art

[0002] In the process of warehouse outbound management, staff are required to pick goods in the warehouse according to the requirements of the orders and ship the picked goods out of the warehouse to achieve the purpose of warehouse outbound delivery. After picking, staff will check the goods, pack them after confirmation, and prepare for transportation. During the outbound delivery process, barcodes need to be pasted, scanned or manually recorded to track the outbound status of the goods. Finally, the goods will be shipped through designated logistics channels to ensure timely and accurate delivery to customers, completing the entire outbound delivery process.

[0003] However, in the actual picking process, workers often face a busy working environment when picking goods, and frequently need to shuttle between different shelves, which makes the picking process appear messy. Due to long-term and high-intensity physical exertion, workers are prone to fatigue and have difficulty concentrating. As fatigue accumulates, the accuracy of picking gradually decreases, resulting in wrong picking, missing picking, or misplacing goods. This not only affects the efficiency of outbound delivery, but may also lead to order errors and customer complaints, thus having a negative impact on the overall operation of the warehouse. Summary of the invention

[0004] The purpose of the present invention is to solve the above-mentioned problems and provide a warehouse intelligent delivery management method and device.

[0005] In a first aspect of the implementation of the present invention, a warehouse intelligent outbound management method is first proposed, the method comprising:

[0006] Obtain the outbound order, and determine the type of goods that need to be shipped out as the target type based on the outbound order information; obtain the images of all goods corresponding to the target type in the warehouse and filter out each of the goods that need to be shipped out as the target goods;

[0007] Obtain the storage height, storage depth, and weight of the target goods, and output the picking difficulty value of the target goods according to the storage height, storage depth, weight, and polynomial regression model;

[0008] A two-dimensional coordinate system is established according to the warehouse, and the coordinates corresponding to each target cargo are obtained. The route is planned based on the warehouse entrance coordinates and the picking difficulty value of the target cargo, the optimal picking path is obtained, and the optimal picking path is sent to the terminal.

[0009] Optionally, acquiring images of all goods corresponding to the target type in the warehouse and selecting each of the goods to be shipped out as the target goods includes:

[0010] Convert the image into a grayscale image and use Gaussian filtering to denoise the image;

[0011] Calculate shape conformity based on the denoised image and the standard cargo template image;

[0012] The texture similarity is calculated based on the pixel value at each position in the standard image and the image corresponding to the target type of goods after denoising;

[0013] Calculate the mean of shape qualification and texture similarity to obtain the goods qualification value; record the goods corresponding to the goods qualification value equal to 1 as qualified goods;

[0014] According to the outbound order information, the quantity of goods corresponding to each target type is obtained. For the qualified goods corresponding to each target type, the time interval from the production date to the current time of each qualified goods is calculated, and the time intervals are sorted in order from large to small. The qualified goods with the corresponding quantity are selected as the target goods set; each of the goods in the target goods set corresponding to all target types is taken as the target goods.

[0015] Optionally, calculating the shape conformity according to the denoised image and the standard image includes:

[0016] The edge features of the denoised image are enhanced by using a sharpening filter; the significant edges in the image are extracted using edge detection methods to form a clear outline;

[0017] Through the results of edge detection, all closed contours in the image are extracted, and the edges are connected using the contour tracking algorithm to generate a complete cargo contour;

[0018] The cargo contour point set is extracted from the image, and the standard cargo template point set is extracted from the standard cargo template image. The two corresponding coordinate points in the two sets of coordinate points are recorded as matching points, and the shape conformity of the target cargo is calculated. The calculation formula is: In the formula, Dq is the shape qualification of the target goods, r is the order number of the matching points, N is the total number of matching points, C 1 (r) and C 2 (r) are the coordinates of the rth matching point in the extracted contour and the standard template, respectively.

[0019] Optionally, outputting the picking difficulty value of the target goods according to the height, depth, weight and polynomial regression model of the goods includes:

[0020] For each target cargo, the storage height, storage depth, and weight of the cargo are obtained, and the storage height, storage depth, and weight are converted into a comprehensive feature vector. The comprehensive feature vector is used as the input of the polynomial regression model. The machine learning model predicts the picking difficulty value label of the target cargo with each set of comprehensive feature vectors as the prediction target, and minimizes the sum of prediction errors of the picking difficulty value labels of all target cargoes as the training target. The machine learning model is trained until the sum of prediction errors converges. The model training is stopped, and the picking difficulty value of the target cargo is determined according to the model output results.

[0021] Optionally, a two-dimensional coordinate system is established according to the warehouse, and the coordinates corresponding to each target item are obtained, and the route is planned in combination with the warehouse entrance coordinates and the picking difficulty of the target item, and the optimal picking path includes:

[0022] A two-dimensional coordinate system is established based on the warehouse, and the coordinates of the warehouse entrance are marked as (0,0); each target item in the warehouse has a unique coordinate, and the target item is marked as i, i∈[1,n], n is the total number; the coordinates of target item i are (x i ,y i );

[0023] The state space is defined as: dp[S][i], where dp[S][i] represents the minimum fatigue factor when the target goods set S has been visited and the target goods i is finally stopped;

[0024] Initialize the fatigue factor from the entrance to each target cargo and set the initial value of the state table. The specific formula is: In the formula, F Entry,i is the fatigue factor from the entrance to the target cargo i, is the coordinate distance from the entrance to the target cargo i, WE i is the picking difficulty value of target goods i;

[0025] For each visited target goods set S and each target goods i, the minimum fatigue factor transferred from a target goods i in the set S to the target goods i is calculated, and the transfer equation of the updated state is as follows:

[0026] dp[S∪{i}][i]=min(dp[S][j]+F j,i )

[0027] In the formula, S is the currently visited target goods set, j is a target goods in the set S, and F j,i is the fatigue factor from target cargo j to target cargo i;

[0028] Traverse all possible visited target goods sets S and each target goods i to update the state table through state transition each time, until it finally contains the results of all visited target goods sets S = {1...n and their minimum fatigue factors;

[0029] When all target goods have been visited, the ultimate goal is to return to the entrance. For each target goods i, the fatigue factor from target goods i back to the entrance is calculated, and the update formula is: dp[{1...n}][i]+F i，Entry ;

[0030] By traversing all possible paths in the state table, the path corresponding to the minimum total fatigue factor is found to be the optimal path.

[0031] In a second aspect of the present invention, a warehouse intelligent outbound management device is provided, the device comprising:

[0032] Target goods module: obtain the outbound order, and determine the type of goods that need to be shipped out as the target type based on the outbound order information; obtain the images of all goods corresponding to the target type in the warehouse and filter out each of the goods that need to be shipped out as the target goods;

[0033] Difficulty value module: obtains the storage height, storage depth, and weight of the target goods, and outputs the picking difficulty value of the target goods based on the storage height, storage depth, weight, and polynomial regression model;

[0034] Picking path management module: establish a two-dimensional coordinate system based on the warehouse, obtain the coordinates corresponding to each target cargo, and plan the route based on the warehouse entrance coordinates and the picking difficulty value of the target cargo to obtain the optimal picking path, and send the optimal picking path to the terminal.

[0035] Optionally, the target cargo module further includes:

[0036] Processing module: convert the image into a grayscale image and use Gaussian filtering to denoise the image;

[0037] Shape conformity module: calculates shape conformity based on the denoised image and the standard cargo template image;

[0038] Texture similarity module: The texture similarity is calculated based on the pixel value of each position in the standard image and the image corresponding to the target type of goods after denoising.

[0039] Qualified goods module: calculate the mean of shape qualification and texture similarity to obtain the qualified value of goods; the goods corresponding to the qualified value equal to 1 are recorded as qualified goods;

[0040] Screening module: Obtain the quantity of goods corresponding to each target type according to the outbound order information. For each target type of qualified goods, calculate the time interval from the production date to the current time of each qualified goods, sort the time intervals from large to small, and select the qualified goods with the corresponding quantity as the target goods set; take each of the goods in the target goods set corresponding to all target types as the target goods.

[0041] Optionally, the shape conformity module comprises:

[0042] Edge extraction module: enhances the edge features of the denoised image through a sharpening filter; uses edge detection methods to extract significant edges in the image to form a clear outline;

[0043] Goods contour module: extracts all closed contours in the image through edge detection results, connects the edges using contour tracking algorithm, and generates a complete goods contour;

[0044] Shape conformity module: extract the cargo contour point set from the image, and extract the standard cargo template point set from the standard cargo template image, record the two corresponding coordinate points in the two sets of coordinate points as matching points, and calculate the shape conformity of the target cargo. The calculation formula is: In the formula, Dq is the shape qualification of the target goods, r is the order number of the matching points, N is the total number of matching points, C 1 (r) and C 2 (r) are the coordinates of the rth matching point in the extracted contour and the standard template, respectively.

[0045] Optionally, the difficulty value module is specifically applied to:

[0046] For each target cargo, the storage height, storage depth, and weight of the cargo are obtained, and the storage height, storage depth, and weight are converted into a comprehensive feature vector. The comprehensive feature vector is used as the input of the polynomial regression model. The machine learning model predicts the picking difficulty value label of the target cargo with each set of comprehensive feature vectors as the prediction target, and minimizes the sum of prediction errors of the picking difficulty value labels of all target cargoes as the training target. The machine learning model is trained until the sum of prediction errors converges. The model training is stopped, and the picking difficulty value of the target cargo is determined according to the model output results.

[0047] Optionally, the picking path management module includes:

[0048] Coordinate system module: Establish a two-dimensional coordinate system based on the warehouse, and mark the coordinate of the warehouse entrance as (0,0); each target item in the warehouse has a unique coordinate, and the target item is marked as i, i∈[1,n], n is the total number; the coordinate of target item i is (xi ,y i );

[0049] State space module: define the state space as: dp[S][i], dp[S][i] represents the minimum fatigue factor when the target goods set S has been visited and the last stop is the target goods i;

[0050] Initialization module: Initialize the fatigue factor from the entrance to each target cargo and set the initial value of the state table. The specific formula is: In the formula, F Entry,i is the fatigue factor from the entrance to the target cargo i, is the coordinate distance from the entrance to the target cargo i, WE i is the picking difficulty value of target goods i;

[0051] Update state module: For each visited target goods set S and each target goods i, calculate the minimum fatigue factor transferred from a target goods i in the set S to the target goods i, and the transfer equation for updating the state is as follows:

[0052] dp[S∪{i}][i]=min(dp[S][j]+F j,i )

[0053] In the formula, S is the currently visited target goods set, j is a target goods in the set S, and F j,i is the fatigue factor from target cargo j to target cargo i;

[0054] Traversal module: traverse all possible visited target goods sets S and each target goods i to update the state table through state transition each time, until it finally contains the result of all visited target goods sets S = {1...n} and their minimum fatigue factors;

[0055] Return module: When all target goods have been visited, the ultimate goal is to return to the entrance. For each target goods i, the fatigue factor from the target goods i back to the entrance is calculated. The update formula is: dp[{1…n}[i]+F i，Entry ;

[0056] Optimal path module: By traversing all possible paths in the state table, the path corresponding to the minimum total fatigue factor is found to be the optimal path.

[0057] Beneficial effects of the present invention:

[0058] 1. The present invention proposes a warehouse intelligent outbound management method and device, which selects target goods in the warehouse according to outbound orders and performs subsequent path planning. The advantage of selecting target goods is that, on the one hand, the picking range can be directly and clearly limited to goods that meet the conditions, which greatly simplifies the task objectives of the staff, so that the staff can directly determine the picking objectives before picking, reduces unnecessary operations, improves the orderliness and accuracy of the picking process, and reduces the fatigue of the pickers. In addition, the goods with earlier production dates are given priority for outbound delivery, which can effectively avoid the quality problems of the inventory goods due to long storage time or exceed the shelf life, and reduce the risk of inventory backlog in the warehouse. In addition, this method also ensures the first-in-first-out (FIFO) principle of goods circulation, and improves the efficiency of inventory management;

[0059] 2. The picking route is planned by taking the location distance between each target product and the storage point and the difficulty of picking each product as fatigue factors, and the staff are asked to pick according to the optimal picking route to ensure that the pickers perform tasks in the warehouse along the route with the minimum fatigue factor, thereby greatly reducing physical exertion and operation time, and avoiding unnecessary repeated walking or excessive walking distance; and the staff does not need to frequently shuttle between different shelves when picking, ensuring the orderly picking process, while also reducing the fatigue of the staff, ensuring the accuracy of picking, reducing the occurrence of wrong picking, missing or misplacing of goods, and reducing the negative impact on the outbound efficiency and the overall operation of the warehouse. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The present invention will be further described below in conjunction with the accompanying drawings.

[0061] Figure 1 It is a flow chart of a warehouse intelligent outbound management method;

[0062] Figure 2 The present invention is a framework diagram of a warehouse intelligent outbound management device. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making any creative work shall fall within the scope of protection of the present invention.

[0065] The embodiment of the present invention provides a warehouse intelligent outbound management method. Figure 1 , Figure 1 A flowchart of a warehouse intelligent outbound management method provided by an embodiment of the present invention. The method comprises the following steps:

[0066] Obtain the outbound order, and determine the type of goods that need to be shipped out as the target type based on the outbound order information; obtain the images of all goods corresponding to the target type in the warehouse and filter out each of the goods that need to be shipped out as the target goods;

[0067] Obtain the storage height, storage depth, and weight of the target goods, and output the picking difficulty value of the target goods according to the storage height, storage depth, weight, and polynomial regression model;

[0068] A two-dimensional coordinate system is established according to the warehouse, and the coordinates corresponding to each target cargo are obtained. The route is planned based on the warehouse entrance coordinates and the picking difficulty value of the target cargo, the optimal picking path is obtained, and the optimal picking path is sent to the terminal.

[0069] Based on an intelligent warehouse outbound management method provided by an embodiment of the present invention, through the above method, when picking goods, the staff does not need to frequently shuttle between different shelves, ensuring the order of the picking process, while also reducing the fatigue of the staff, ensuring the accuracy of picking, reducing the occurrence of wrong picking, missing or misplacing of goods, and reducing the negative impact on outbound efficiency and the overall operation of the warehouse.

[0070] It should be noted that obtaining the outbound order and determining the type of goods that need to be shipped out based on the outbound order information as the target type, this step mainly clarifies the specific requirements of the outbound task to ensure that the subsequent picking, checking and outbound processes can be carried out efficiently and in a targeted manner. This step helps warehouse staff or systems narrow the search scope by identifying the target type, thereby improving operational efficiency and reducing the error rate; for example: a warehouse receives an outbound order with the following content: target goods A and the quantity is n; target goods B and the quantity is m; target goods C and the quantity is k. The system identifies the target types that need to be shipped out based on the order information, namely target goods A, target goods B and target goods C. Subsequently, the system locates the storage location of the corresponding goods in the warehouse based on the inventory data, and screens out inventory items that meet the specifications.

[0071] In one embodiment, images of all goods corresponding to the target type in the warehouse are obtained, and all qualified goods are screened out according to the images, including:

[0072] Convert each image to grayscale, retain the brightness information, and reduce the data dimension.

[0073] Use Gaussian filtering to remove image noise and make the outline clearer;

[0074] Enhance the edge features of the image through a sharpening filter to highlight the contours of the goods;

[0075] Use edge detection methods (such as Canny edge detection) to extract significant edges in the image and form clear contours;

[0076] Through the results of edge detection, all closed contours in the image are extracted, and the edges are connected using a contour tracking algorithm (such as a pixel gradient-based method) to generate a complete cargo contour;

[0077] A cargo contour point set extracted from the image (obtained by sampling or interpolation after contour extraction), and a standard cargo template point set extracted from the standard cargo template image (with the same sampling order and number of points), and the two sets of point coordinates are normalized to eliminate the influence of shooting angle or scale difference on shape matching (for example, the coordinate value is divided by the maximum width or height of the object contour);

[0078] The two corresponding coordinate points in the two sets of coordinate points are recorded as matching points, and the shape conformity of the target goods is calculated. The calculation formula is: In the formula, Dq is the shape qualification of the target goods, r is the order number of the matching points, N is the total number of matching points, C 1 (r) and C 2 (r) are the coordinates of the rth matching point in the extracted contour and the standard template, respectively.

[0079] It should be noted that in the above steps, each image represents a cargo image corresponding to a target type. A high-resolution camera can be installed above or around the shelf to scan the cargo on the shelf, update the cargo image data, and filter out the images of all cargo corresponding to the target type. Other acquisition methods are also possible and are not limited. The standard cargo template image is also a standard cargo template image of the corresponding type of cargo stored in the warehouse in advance, which is determined specifically according to the type of cargo and is not limited or elaborated.

[0080] It should be noted that the above steps are illustrated by an example, assuming that the target cargo is a standard rectangular box, and the standard cargo template point set is as follows (after normalization): {(0,0), (1,0), (1,1), (0,1)};

[0081] The target cargo contour point set extracted from the image is: {(0.05, 0.02), (0.98, -0.01), (1.01, 0.97), (0.02, 1.03)};

[0082] The coordinate differences between the matching points are: {(0.05, 0.02), (-0.02, -0.01), (0.01, -0.03), (0.02, 0.03)}, so the shape conformity of the goods is

[0083] It should be noted that the shape conformity is used to measure the actual outline of the goods corresponding to the target type and the outline of the standard goods, that is, the shape similarity, because the shape similarity reflects whether the size of the goods meets expectations; by comparing the coordinate differences between the target goods and the standard goods template at each matching point, the length-width ratio, edge features and overall geometric shape of the target object can be evaluated. This similarity calculation method can help us determine whether the goods have dimensional anomalies caused by deformation, damage or improper storage, such as excessive compression, expansion or deformation; the value range of the shape conformity is between 0 and 1. The closer it is to 1, the higher the degree of match between the size of the goods and the standard template. Conversely, it may indicate that the goods have size deviations or shape damage; in this way, not only can the appearance of the goods be ensured to meet the standards, but also unqualified or damaged items can be effectively screened out, thereby realizing the quality inspection and management of warehouse storage items. When the shape conformity of the goods corresponding to the target type is greater, it means that the quality of the goods is more qualified and can be used as goods that meet the conditions for leaving the warehouse.

[0084] In one implementation method, the above method uses contour matching to calculate the shape conformity of the goods corresponding to the target type. The advantage is that it can effectively solve the shape recognition problems caused by shooting angles, scale differences or changes in object posture. The contour matching method does not rely on color or texture features, but compares the edges and geometric shapes of objects, so that it can still maintain a high accuracy under different environments and conditions. This can improve the robustness of goods identification, reduce misidentification caused by external interference, and ensure that the warehouse management system can operate stably and efficiently in a complex environment.

[0085] In one embodiment, acquiring images of all goods corresponding to the target type in the warehouse and screening out all qualified goods according to the images further includes:

[0086] Convert the image to grayscale to reduce the amount of calculation;

[0087] Gaussian filtering is used to denoise the image to reduce the impact of noise on the template matching results.

[0088] If there are illumination changes in the image, brightness normalization or contrast enhancement can be performed to reduce the impact of changing illumination conditions;

[0089] Calculate the similarity between the standard image and the image corresponding to the target type of goods, and record the similarity as texture similarity. The formula is:

[0090]

[0091] Where NCC is the texture similarity, I(x,v) is the pixel value of the image position (x,v) corresponding to the target type of goods, T(x,v) is the pixel value of the standard image position (x,v), and μ I and μ T are the average pixel values ​​of the image corresponding to the target type of goods and the standard image;

[0092] Calculate the mean of shape conformity and texture similarity to obtain the goods conformity value; record the goods corresponding to the goods conformity value equal to 1 as conforming goods.

[0093] It should be noted that the standard image refers to the standard image of the corresponding type of goods stored in the warehouse in advance, which is determined according to the type of goods and is not limited or elaborated on.

[0094] It should be noted that texture similarity is used to measure the texture similarity between the image corresponding to the target type of goods and the standard image, because texture similarity reflects whether the goods have been damaged. If the surface of the goods is intact and there are no obvious scratches, dents or cracks, its texture features should be highly consistent with the texture features of the standard template image; therefore, by comparing the texture similarity between the target image and the standard image, it is possible to determine whether any abnormalities have occurred on the surface of the goods, such as damage or defects; and the value range of texture similarity is between 0 and 1. The closer it is to 0, the more deformation or damage there is on the surface of the goods, while a high similarity means that the surface of the goods has maintained its original characteristics and state, indicating that the quality of the goods is relatively qualified and can be used as goods that meet the conditions for leaving the warehouse.

[0095] In one implementation, the above method uses pixel values ​​to calculate the similarity of cargo textures corresponding to the target type. The advantage is that it can directly reflect the detail changes at each location in the image, especially when there are slight changes on the surface of the object. For example, slight scratches, stains or other imperceptible damage on the surface often lead to differences in local pixel values. By comparing the pixel values ​​between the target image and the standard image pixel by pixel, these differences in details can be captured more accurately, thereby improving the accuracy of the judgment of the surface integrity of the cargo. In addition, this method is relatively simple, does not require a complex feature extraction process, can efficiently process large-scale image data, and is suitable for real-time monitoring and automated detection systems.

[0096] In one embodiment, selecting each of the goods that need to be shipped out as target goods according to all the qualified goods includes:

[0097] According to the outbound order information, the quantity of goods corresponding to each target type is obtained, and each of the goods that need to be shipped out is selected as the target goods from the qualified goods corresponding to each target type. Specifically:

[0098] For each qualified product corresponding to the target type, calculate the time interval from the production date of each qualified product to the current time, sort the time intervals from large to small, and select the qualified products with the corresponding quantity as the target product set;

[0099] Each cargo in the target cargo set corresponding to all target types is taken as the target cargo.

[0100] In one implementation method, through the above method, goods with earlier production dates are given priority for shipment, which can effectively avoid quality problems or expiration of the shelf life of inventory goods due to long storage time, and reduce the risk of inventory backlog in the warehouse. In addition, this method also ensures the first-in-first-out (FIFO) principle of goods circulation, thereby improving the efficiency of inventory management; at the same time, by accurately matching the shipment order requirements, it can avoid unnecessary picking operations, optimize the picking process, save manpower and time costs, and thus improve the overall warehouse operation efficiency and customer satisfaction.

[0101] It should be noted that the benefit of screening out the target goods is that the picking scope can be clearly limited to the qualified goods, which greatly simplifies the task objectives of the staff; this method can effectively reduce the ineffective shuttle between the shelves of the staff, thereby reducing the speed of physical exertion and fatigue accumulation; in a busy working environment, this precise screening method helps to improve the orderliness and accuracy of the picking process, avoid the problem of wrong picking, missing or misplacing due to lack of concentration, and fundamentally reduce the possibility of order errors and customer complaints; at the same time, this also improves the efficiency of outbound delivery, ensures that the warehouse can operate smoothly and maintain a good customer experience; in addition, in the subsequent planning of the picking path, the target goods are directly marked, and the path is planned according to the location of the target goods in the warehouse, which reduces the staff's aimless movement and repeated walking in the warehouse. This method can significantly shorten the picking time, improve the picking efficiency, reduce physical exertion, and enable the staff to focus more on completing the picking task accurately. By marking the target goods and planning a reasonable path, the picking path can be optimized, avoiding detours or omissions caused by unreasonable paths, thereby further improving work efficiency and the overall effect of warehouse operations. This planning method can also improve the degree of automation in warehouses and provide support for the implementation of intelligent warehousing systems.

[0102] In one embodiment, outputting the picking difficulty value of the target goods according to the height, depth, weight and polynomial regression model of the goods includes:

[0103] For each target cargo, the storage height, storage depth, and weight of the cargo are obtained, and the storage height, storage depth, and weight are converted into a comprehensive feature vector. The comprehensive feature vector is used as the input of the polynomial regression model. The machine learning model predicts the picking difficulty value label of the target cargo with each set of comprehensive feature vectors as the prediction target, and minimizes the sum of prediction errors of the picking difficulty value labels of all target cargoes as the training target. The machine learning model is trained until the sum of prediction errors converges. The model training is stopped, and the picking difficulty value of the target cargo is determined according to the model output results.

[0104] The method for obtaining the picking difficulty value of the target goods is as follows: from the comprehensive feature vector training data of the trained machine learning model, the corresponding function expression is obtained: WE=F(TY, DS, UK); wherein F is the output function of the model, TY, DS, UK are the storage height, storage depth, and weight of the goods, respectively, and WE is the picking difficulty value of the target goods.

[0105] It should be noted that the height, depth and weight of the goods can be obtained through the warehouse management system (WMS) or the barcode / RFID technology inside the warehouse. When the goods are put into storage, the warehouse system will record the specific storage location of each item, including the height (such as the number of shelf layers), storage depth (the location of the goods on the shelf), and weight (usually measured by the weighing equipment in the warehouse system or input at the time of storage). This data can be stored and used in the subsequent picking process. By combining with other warehouse management information, it provides accurate target goods feature data for machine learning model training and prediction of picking difficulty values.

[0106] In one embodiment, a two-dimensional coordinate system is established according to the warehouse, and the coordinates corresponding to each target item are obtained. The route is planned in combination with the warehouse entrance coordinates and the difficulty of picking the target item. The optimal picking path includes:

[0107] A two-dimensional coordinate system is established based on the warehouse, and the coordinates of the warehouse entrance are marked as (0,0); each target item in the warehouse has a unique coordinate, and the target item is marked as i, i∈[1,n], n is the total number; the coordinates of target item i are (x i ,y i );

[0108] In dynamic programming, the state space can be defined as the minimum fatigue factor of the target goods set that have been visited and the last stop at a target goods. The state space is defined as: dp[S][i], dp[S][i] represents the minimum fatigue factor when the target goods set S has been visited and the last stop is the target goods i;

[0109] Initialize the fatigue factor from the entrance to each target cargo and set the initial value of the state table:

[0110] Calculate the fatigue factor from the warehouse entrance (0,0) to each target item i. This fatigue factor is calculated based on the distance and the picking difficulty value. The specific calculation formula is:

[0111] In the formula, F Entry,i From the entrance to the destination i The fatigue factor, For import to target goods i Coordinate distance, WE i Target cargo i The picking difficulty value;

[0112] Therefore, the initial state is: dp[{i}][i] = F Entry,i ;

[0113] For each visited target goods set S and each target goods i, it is necessary to calculate the minimum fatigue factor transferred from a target goods i in the set S to the target goods i, and the transfer equation for updating the state is as follows:

[0114] dp[S∪{i}][i]=min(dp[S][j]+F j,i )

[0115] In the formula, S is the currently visited target goods set, j is a target goods in the set S, and F j,i is the fatigue factor from target cargo j to target cargo i;

[0116] Each time the state table is updated, the minimum fatigue factor from a target item j to the current target item i is calculated based on all the target items in the visited target item set S, ensuring that the minimum fatigue factor is selected each time the state table is updated;

[0117] Traverse all possible visited target goods sets S and each target goods i to update the state table through state transition each time, until it finally contains the result of all visited target goods sets S = {1...n} and their minimum fatigue factors;

[0118] For each possible set of target goods, calculate the minimum fatigue factor transferred from each target goods j in the set S to the current target goods i;

[0119] When all target goods have been visited, the ultimate goal is to return to the entrance. For each target goods i, the fatigue factor from target goods i back to the entrance is calculated, and the update formula is: dp[{1...n}][i]+F i，Entry ;

[0120] By traversing all possible paths in the state table, the minimum total fatigue factor is found, that is, the minimum fatigue factor is: min(dp[{1…n}][1]+F 1，Entry , dp[{1…n}][2]+F 2，Entry .......dp[{1……n}[n]+F n，Entry );

[0121] Among them, the path corresponding to the smallest fatigue factor is the optimal path.

[0122] It should be noted that the above steps are briefly described by examples, for example:

[0123] Assume that in a warehouse, the entrance coordinates are (0,0), the number of target goods is 3, and the coordinates and picking difficulty values ​​of the target goods are as follows:

[0124] Target cargo number i <![CDATA[Coordinates (x i , y i )]]> <![CDATA[Picking difficulty value WE i > 1 (3,4) 2.0 2 (6,8) 1.5 3 (9,12) 1.8

[0125] 1. Initialize the status table;

[0126] Calculate the fatigue factor from the entrance to each cargo:

[0127] For target cargo 1:

[0128] For target cargo 2:

[0129] For target cargo 3:

[0130] Initial state table: dp[{1}][1]=3.5; dp[{2}][2]=5.75; dp[{3}][3]=8.4;

[0131] 2. State transfer:

[0132] Now, calculate the minimum fatigue factor for transferring from a certain set of goods S to another target goods.

[0133] For set S = {1}, transfer from goods 1 to goods 2:

[0134] After the transfer: dp[{1, 2}][2] = F Entry,1 +3.25=3.5+3.25=6.75;

[0135] For set S = {1}, transfer from item 1 to item 3:

[0136] After the transfer, the state is: dp[{1, 3}][3] = F Entry ,1+5.9=3.5+5.9=9.4;

[0137] And so on, calculate all possible state transitions: from S = {2} to goods 1 and goods 3; from S = {3} to goods 1 and goods 2;

[0138] 3. Complete all state transfers: gradually expand the visited cargo set until it includes all cargo sets S = {1, 2, 3}, and transfer from S = {1, 2} to cargo 3: calculate the fatigue factor from cargo 1 and cargo 2 to cargo 3, and select the minimum value:

[0139] dp[{1,2,3}][3]=min(dp[1,2][2]+F 2,3 ,dp[1,2][1]+F 1,3 )

[0140] in, F 1,3 = dp[{1,3}][3] = 9.4; transfer from S = {1,3} to goods 2, similar calculation;

[0141] 4. Return to the warehouse entrance;

[0142] When all goods have been visited, finally return to the entrance:

[0143] Return entrance from cargo 1:

[0144] Return entrance from Cargo 2:

[0145] Return entrance from cargo 3:

[0146] Update the final state table, calculate the total fatigue factor, and select the minimum value:

[0147] min(dp[{1,2,3}][1]+F 1,Entry ,dp[{1,2,3}][1]+F 1,Entry ,dp[{1,2,3}][1]+F 1,Entry );

[0148] 5. Optimal path: Determine the specific path by inverting the selection in the state table; for example:

[0149] The path is: entrance (0,0) → cargo 1 (3,4) → cargo 2 (6,8) → cargo 3 (9,12) → entrance, and the minimum fatigue factor is 20.15.

[0150] This process dynamically plans the path by combining distance and picking difficulty values, gradually updates the state table, and finally finds the optimal path to minimize the fatigue factor. This method improves picking efficiency, reduces physical exertion, and is suitable for intelligent warehouse management.

[0151] It should be noted that the benefits of obtaining the optimal picking path through the above method are mainly reflected in improving warehouse operation efficiency, reducing fatigue and optimizing resource allocation. First, the optimal path planning can ensure that the pickers perform tasks in the warehouse along the route with the minimum fatigue factor, thereby greatly reducing physical exertion and operation time, and avoiding unnecessary repeated walking or excessive walking distance; secondly, by considering the picking difficulty of each target goods and the distance between each target goods, the waste of warehouse resources can be minimized and the overall operational efficiency can be improved; in addition, the optimal path can also reduce errors or delays caused by fatigue, thereby improving picking accuracy and order processing speed, and further improving customer experience. In short, the optimal picking path obtained by dynamic planning not only helps to reduce operating costs, but also improves the flexibility and responsiveness of warehouse management.

[0152] Based on the same inventive concept, the embodiment of the present invention also provides a warehouse intelligent outbound management device. Figure 2 , Figure 2 A framework diagram of a warehouse intelligent outbound management device provided by an embodiment of the present invention, the device comprising:

[0153] Target goods module: obtain the outbound order, and determine the type of goods that need to be shipped out as the target type based on the outbound order information; obtain the images of all goods corresponding to the target type in the warehouse and filter out each of the goods that need to be shipped out as the target goods;

[0154] Difficulty value module: obtains the storage height, storage depth, and weight of the target goods, and outputs the picking difficulty value of the target goods based on the storage height, storage depth, weight, and polynomial regression model;

[0155] Picking path management module: establish a two-dimensional coordinate system based on the warehouse, obtain the coordinates corresponding to each target cargo, and plan the route based on the warehouse entrance coordinates and the picking difficulty value of the target cargo to obtain the optimal picking path, and send the optimal picking path to the terminal.

[0156] Based on the intelligent warehouse outbound management device provided by the embodiment of the present invention, through the above method, the staff does not need to frequently shuttle between different shelves when picking goods, ensuring the order of the picking process, while also reducing the fatigue of the staff, ensuring the accuracy of picking, reducing the occurrence of wrong picking, missing or misplacing of goods, and reducing the negative impact on outbound efficiency and the overall operation of the warehouse.

[0157] In one embodiment, the target cargo module further comprises:

[0158] Processing module: convert the image into a grayscale image and use Gaussian filtering to denoise the image;

[0159] Shape conformity module: calculates shape conformity based on the denoised image and the standard cargo template image;

[0160] Texture similarity module: The texture similarity is calculated based on the pixel value of each position in the standard image and the image corresponding to the target type of goods after denoising.

[0161] Qualified goods module: calculate the mean of shape qualification and texture similarity to obtain the qualified value of goods; the goods corresponding to the qualified value equal to 1 are recorded as qualified goods;

[0162] Screening module: Obtain the quantity of goods corresponding to each target type according to the outbound order information. For each target type of qualified goods, calculate the time interval from the production date to the current time of each qualified goods, sort the time intervals from large to small, and select the qualified goods with the corresponding quantity as the target goods set; take each of the goods in the target goods set corresponding to all target types as the target goods.

[0163] In one embodiment, the shape conformity module comprises:

[0164] Edge extraction module: enhances the edge features of the denoised image through a sharpening filter; uses edge detection methods to extract significant edges in the image to form a clear outline;

[0165] Goods contour module: extracts all closed contours in the image through edge detection results, connects the edges using contour tracking algorithm, and generates a complete goods contour;

[0166] Shape conformity module: extract the cargo contour point set from the image, and extract the standard cargo template point set from the standard cargo template image, record the two corresponding coordinate points in the two sets of coordinate points as matching points, and calculate the shape conformity of the target cargo. The calculation formula is: In the formula, Dq is the shape qualification of the target goods, r is the order number of the matching points, N is the total number of matching points, C 1 (r) and C 2 (r) are the coordinates of the rth matching point in the extracted contour and the standard template, respectively.

[0167] In one embodiment, the difficulty value module is specifically applied to:

[0168] For each target cargo, the storage height, storage depth, and weight of the cargo are obtained, and the storage height, storage depth, and weight are converted into a comprehensive feature vector. The comprehensive feature vector is used as the input of the polynomial regression model. The machine learning model predicts the picking difficulty value label of the target cargo with each set of comprehensive feature vectors as the prediction target, and minimizes the sum of prediction errors of the picking difficulty value labels of all target cargoes as the training target. The machine learning model is trained until the sum of prediction errors converges. The model training is stopped, and the picking difficulty value of the target cargo is determined according to the model output results.

[0169] In one embodiment, the picking path management module includes:

[0170] Coordinate system module: Establish a two-dimensional coordinate system based on the warehouse, and mark the coordinate of the warehouse entrance as (0,0); each target item in the warehouse has a unique coordinate, and the target item is marked as i, i∈[1,n], n is the total number; the coordinate of target item i is (x i ,y i );

[0171] State space module: define the state space as: dp[S][i], dp[S][i] represents the minimum fatigue factor when the target goods set S has been visited and the last stop is the target goods i;

[0172] Initialization module: Initialize the fatigue factor from the entrance to each target cargo and set the initial value of the state table. The specific formula is: In the formula, F Entry,i is the fatigue factor from the entrance to the target cargo i, is the coordinate distance from the entrance to the target cargo i, WE i is the picking difficulty value of target goods i;

[0173] Update state module: For each visited target goods set S and each target goods i, calculate the minimum fatigue factor transferred from a target goods i in the set S to the target goods i, and the transfer equation for updating the state is as follows:

[0174] dp[S∪{i}][i]=min(dp[S][j]+F j,i )

[0175] In the formula, S is the currently visited target goods set, j is a target goods in the set S, and F j,i is the fatigue factor from target cargo j to target cargo i;

[0176] Traversal module: traverse all possible visited target goods sets S and each target goods i to update the state table through state transition each time, until it finally contains the result of all visited target goods sets S = {1...n} and their minimum fatigue factors;

[0177] Return module: When all target goods have been visited, the ultimate goal is to return to the entrance. For each target goods i, the fatigue factor from target goods i back to the entrance is calculated. The update formula is: dp[{1…n}][i]+F i，Entry ;

[0178] Optimal path module: By traversing all possible paths in the state table, the path corresponding to the minimum total fatigue factor is found to be the optimal path.

[0179] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be used to artificially limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A warehouse intelligent outbound management method, characterized in that: The following steps are involved: Obtain the outbound order, and determine the type of goods that need to be shipped out as the target type based on the outbound order information; obtain the images of all goods corresponding to the target type in the warehouse and filter out each of the goods that need to be shipped out as the target goods; Obtain the storage height, storage depth, and weight of the target goods, and output the picking difficulty value of the target goods according to the storage height, storage depth, weight, and polynomial regression model; A two-dimensional coordinate system is established according to the warehouse, and the coordinates corresponding to each target cargo are obtained. The route is planned based on the warehouse entrance coordinates and the picking difficulty value of the target cargo, the optimal picking path is obtained, and the optimal picking path is sent to the terminal.

2. The intelligent warehouse outbound management method according to claim 1 is characterized in that: Get the images of all goods corresponding to the target type in the warehouse and filter out each goods that need to be shipped out as target goods, including: Convert the image into a grayscale image and use Gaussian filtering to denoise the image; Calculate shape conformity based on the denoised image and the standard cargo template image; The texture similarity is calculated based on the pixel value at each position in the standard image and the image corresponding to the target type of goods after denoising; Calculate the mean of shape qualification and texture similarity to obtain the goods qualification value; record the goods corresponding to the goods qualification value equal to 1 as qualified goods; According to the outbound order information, the quantity of goods corresponding to each target type is obtained. For the qualified goods corresponding to each target type, the time interval from the production date to the current time of each qualified goods is calculated, and the time intervals are sorted in order from large to small. The qualified goods with the corresponding quantity are selected as the target goods set; each of the goods in the target goods set corresponding to all target types is taken as the target goods.

3. A warehouse intelligent outbound management method according to claim 2, characterized in that: Calculating the shape conformity based on the denoised image and the standard image includes: The edge features of the denoised image are enhanced by using a sharpening filter; the significant edges in the image are extracted using edge detection methods to form a clear outline; Through the results of edge detection, all closed contours in the image are extracted, and the edges are connected using the contour tracking algorithm to generate a complete cargo contour; The cargo contour point set is extracted from the image, and the standard cargo template point set is extracted from the standard cargo template image. The two corresponding coordinate points in the two sets of coordinate points are recorded as matching points, and the shape conformity of the target cargo is calculated. The calculation formula is: Where Dq is the shape conformity of the target goods, r is the order number of the matching point, N is the total number of matching points, C1(r) and C2(r) are the coordinates of the rth matching point in the extracted contour and the standard template, respectively.

4. The intelligent warehouse outbound management method according to claim 1 is characterized in that: The picking difficulty values ​​of the target goods are output based on the height, depth, weight and polynomial regression model of the goods, including: For each target cargo, the storage height, storage depth, and weight of the cargo are obtained, and the storage height, storage depth, and weight are converted into a comprehensive feature vector. The comprehensive feature vector is used as the input of the polynomial regression model. The machine learning model predicts the picking difficulty value label of the target cargo with each set of comprehensive feature vectors as the prediction target, and minimizes the sum of prediction errors of the picking difficulty value labels of all target cargoes as the training target. The machine learning model is trained until the sum of prediction errors converges. The model training is stopped, and the picking difficulty value of the target cargo is determined according to the model output results.

5. The intelligent warehouse outbound management method according to claim 1 is characterized in that: A two-dimensional coordinate system is established based on the warehouse, and the coordinates corresponding to each target item are obtained. The route is planned based on the warehouse entrance coordinates and the picking difficulty of the target item. The optimal picking path includes: A two-dimensional coordinate system is established based on the warehouse, and the coordinates of the warehouse entrance are marked as (0,0); each target item in the warehouse has a unique coordinate, and the target item is marked as i, i∈[1,n], n is the total number; the coordinates of target item i are (x i ,y i ); The state space is defined as: dp[S][i], where dp[S][i] represents the minimum fatigue factor when the target goods set S has been visited and the target goods i is finally stopped; Initialize the fatigue factor from the entrance to each target cargo and set the initial value of the state table. The specific formula is: In the formula, F Entry,i is the fatigue factor from the entrance to the target cargo i, is the coordinate distance from the entrance to the target cargo i, WE i is the picking difficulty value of target goods i; For each visited target goods set S and each target goods i, the minimum fatigue factor transferred from a target goods i in the set S to the target goods i is calculated, and the transfer equation of the updated state is as follows: dp[S∪{i}[i]=min(dp[S][j]+F j,i ) In the formula, S is the currently visited target goods set, j is a target goods in the set S, and F j,i is the fatigue factor from target cargo j to target cargo i; Traverse all possible visited target goods sets S and each target goods i to update the state table through state transition each time, until it finally contains the result of all visited target goods sets S = {1...n} and their minimum fatigue factors; When all target goods have been visited, the ultimate goal is to return to the entrance. For each target goods i, the fatigue factor from target goods i back to the entrance is calculated, and the update formula is: dp[{1...n}][i]+F i，Entry ; By traversing all possible paths in the state table, the path corresponding to the minimum total fatigue factor is found to be the optimal path.

6. A warehouse intelligent outbound management device, characterized in that: The device comprises: Target goods module: obtain the outbound order, and determine the type of goods that need to be shipped out as the target type based on the outbound order information; obtain the images of all goods corresponding to the target type in the warehouse and filter out each of the goods that need to be shipped out as the target goods; Difficulty value module: obtains the storage height, storage depth, and weight of the target goods, and outputs the picking difficulty value of the target goods based on the storage height, storage depth, weight, and polynomial regression model; Picking path management module: establish a two-dimensional coordinate system based on the warehouse, obtain the coordinates corresponding to each target cargo, and plan the route based on the warehouse entrance coordinates and the picking difficulty value of the target cargo to obtain the optimal picking path, and send the optimal picking path to the terminal.

7. The intelligent warehouse outbound management device according to claim 6 is characterized in that: The target cargo module also includes: Processing module: convert the image into a grayscale image and use Gaussian filtering to denoise the image; Shape conformity module: calculates shape conformity based on the denoised image and the standard cargo template image; Texture similarity module: The texture similarity is calculated based on the pixel value of each position in the standard image and the image corresponding to the target type of goods after denoising. Qualified goods module: calculate the mean of shape qualification and texture similarity to obtain the qualified value of goods; the goods corresponding to the qualified value equal to 1 are recorded as qualified goods; Screening module: Obtain the quantity of goods corresponding to each target type according to the outbound order information. For each target type of qualified goods, calculate the time interval from the production date to the current time of each qualified goods, sort the time intervals from large to small, and select the qualified goods with the corresponding quantity as the target goods set; take each of the goods in the target goods set corresponding to all target types as the target goods.

8. The intelligent warehouse outbound management device according to claim 7 is characterized in that: The shape conformity module includes: Edge extraction module: enhances the edge features of the denoised image through a sharpening filter; uses edge detection methods to extract significant edges in the image to form a clear outline; Goods contour module: extracts all closed contours in the image through edge detection results, connects the edges using contour tracking algorithm, and generates a complete goods contour; Shape conformity module: extract the cargo contour point set from the image, and extract the standard cargo template point set from the standard cargo template image, record the two corresponding coordinate points in the two sets of coordinate points as matching points, and calculate the shape conformity of the target cargo. The calculation formula is: Where Dq is the shape conformity of the target goods, r is the order number of the matching point, N is the total number of matching points, C1(r) and C2(r) are the coordinates of the rth matching point in the extracted contour and the standard template, respectively.

9. The intelligent warehouse outbound management device according to claim 6, characterized in that: The difficulty value module is specifically used for: For each target cargo, the storage height, storage depth, and weight of the cargo are obtained, and the storage height, storage depth, and weight are converted into a comprehensive feature vector. The comprehensive feature vector is used as the input of the polynomial regression model. The machine learning model predicts the picking difficulty value label of the target cargo with each set of comprehensive feature vectors as the prediction target, and minimizes the sum of prediction errors of the picking difficulty value labels of all target cargoes as the training target. The machine learning model is trained until the sum of prediction errors converges. The model training is stopped, and the picking difficulty value of the target cargo is determined according to the model output results.

10. The intelligent warehouse outbound management device according to claim 6, characterized in that: The picking path management module includes: Coordinate system module: Establish a two-dimensional coordinate system based on the warehouse, and mark the coordinate of the warehouse entrance as (0,0); each target item in the warehouse has a unique coordinate, and the target item is marked as i, i∈[1,n], n is the total number; the coordinate of target item i is (x i ,y i ); State space module: define the state space as: dp[S][i], dp[S][i] represents the minimum fatigue factor when the target goods set S has been visited and the last stop is the target goods i; Initialization module: Initialize the fatigue factor from the entrance to each target cargo and set the initial value of the state table. The specific formula is: In the formula, F Entry,i is the fatigue factor from the entrance to the target cargo i, is the coordinate distance from the entrance to the target cargo i, WE i is the picking difficulty value of target goods i; Update state module: For each visited target goods set S and each target goods i, calculate the minimum fatigue factor transferred from a target goods i in the set S to the target goods i, and the transfer equation for updating the state is as follows: dp[S∪{i}][i]=min(dp[S][j]+F j,i ) In the formula, S is the currently visited target goods set, j is a target goods in the set S, and F j,i is the fatigue factor from target cargo j to target cargo i; Traversal module: traverse all possible visited target goods sets S and each target goods i to update the state table through state transition each time, until it finally contains the result of all visited target goods sets S = {1...n} and their minimum fatigue factors; Return module: When all target goods have been visited, the ultimate goal is to return to the entrance. For each target goods i, the fatigue factor from target goods i back to the entrance is calculated. The update formula is: dp[{1…n}][i]+F i，Entry ; Optimal path module: By traversing all possible paths in the state table, the path corresponding to the minimum total fatigue factor is found to be the optimal path.

Images