Cargo identification and sorting methods for intelligent warehousing

Abstract

The present invention discloses a method for identifying and sorting goods in intelligent warehousing, and relates to the technical field of goods sorting. The method includes: interacting with the storage management module of the target warehouse to obtain a pool of goods to be sorted and a list of goods to be stored; performing feature identification on the pool of goods to be sorted to determine first goods feature information; performing warehousing authentication to obtain a first goods warehousing authentication result; performing multi-level joint optimization on the sorting control decision space to determine a first goods sorting control strategy; based on the sorting robot, sorting the first goods to be sorted according to the first goods sorting control strategy. The present invention solves the technical problems in the prior art that the accuracy and efficiency of goods feature identification in goods identification and sorting are low, and it is difficult to balance the sorting efficiency and loss, and achieves the technical effect of accurately identifying goods features, efficient warehousing authentication, and determining the optimal sorting control strategy, thereby accurately and balancing the sorting efficiency and loss.

Description

Technical Field

[0001] The present invention relates to the technical field of cargo sorting, and in particular to a cargo identification and sorting method for intelligent warehousing. Background Art

[0002] With the rapid development of the logistics industry, intelligent warehousing has become a key link in improving logistics efficiency and reducing costs. In the intelligent warehousing system, the identification and sorting of goods is one of the core tasks. Traditional methods of sorting goods in warehouses often rely on manual operations, which are inefficient and prone to errors. With the advancement of technology, automated sorting systems have gradually emerged, but the existing sorting systems still have some problems when facing complex and diverse types of goods and changing warehousing needs. In terms of cargo feature identification, the accuracy and efficiency need to be improved. Inaccurate feature identification may lead to errors in cargo entry authentication and affect the subsequent sorting process. At the same time, the balance between sorting efficiency and sorting losses is also a difficult problem. Simply pursuing high sorting efficiency may lead to increased sorting losses such as damage and loss of goods, while focusing too much on reducing sorting losses may reduce sorting efficiency and affect the overall operational efficiency of warehousing.

[0003] The existing technology has technical problems such as low accuracy and efficiency in cargo feature recognition during cargo identification and sorting, and difficulty in balancing sorting efficiency and losses. Summary of the invention

[0004] The present application provides a method for cargo identification and sorting in intelligent warehousing, which is used to solve the technical problems in the prior art of cargo identification and sorting, such as low accuracy and efficiency in cargo feature identification, and difficulty in balancing sorting efficiency and losses.

[0005] In view of the above problems, the present application provides a method for identifying and sorting goods in intelligent warehousing, the method comprising:

[0006] The storage management module of the interactive target warehouse obtains a pool of goods to be sorted and a list of goods to be stored; performs feature recognition on the pool of goods to be sorted according to the Hu-Rajan-Chia model to determine first goods feature information corresponding to the first goods to be sorted; based on the first goods feature information, performs warehousing authentication on the first goods to be sorted according to the list of goods to be stored to obtain a first goods warehousing authentication result; when the first goods warehousing authentication result is passed, reads the first goods real-time position and the first goods storage position corresponding to the first goods to be sorted, and performs control parameter decision on the sorting robot in combination with the first goods feature information to obtain a sorting control decision space; based on the first goods sorting loss threshold and the first goods expected sorting efficiency corresponding to the first goods to be sorted, performs multi-level joint optimization on the sorting control decision space to determine a first goods sorting control strategy; based on the sorting robot, sorts the first goods to be sorted according to the first goods sorting control strategy.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] The storage management module of the interactive target warehouse obtains the pool of goods to be sorted and the list of goods to be stored; the pool of goods to be sorted is characterized by the Hu-Rajan-Chia model to determine the first goods characteristic information corresponding to the first goods to be sorted; the first goods to be sorted are authenticated for entry into the warehouse to obtain the first goods entry authentication result; when the first goods entry authentication result is passed, the real-time location of the first goods and the storage location of the first goods are read to obtain the sorting control decision space; the sorting control decision space is optimized at multiple levels to determine the sorting control strategy for the first goods; based on the sorting robot, the first goods to be sorted are sorted according to the first goods sorting control strategy. The technical effect of accurately identifying goods characteristics, efficient entry authentication, and determining the optimal sorting control strategy is achieved, thereby accurately and balancing sorting efficiency and loss. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A flow chart of a method for identifying and sorting goods in intelligent warehousing provided in an embodiment of the present application;

[0011] Figure 2A schematic diagram of a process for obtaining a first cargo entry authentication result in the cargo identification and sorting method for intelligent warehousing provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The present application provides a method for cargo identification and sorting in intelligent warehousing, which is used to solve the technical problems in the prior art of cargo identification and sorting, such as low accuracy and efficiency in cargo feature identification, and difficulty in balancing sorting efficiency and losses.

[0013] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0014] Examples, such as Figure 1 As shown, the present application provides a method for identifying and sorting goods in intelligent warehousing, the method comprising:

[0015] Step S100: Interact with the storage management module of the target warehouse to obtain a pool of goods to be sorted and a list of goods to be stored.

[0016] Specifically, first, a connection is established with the warehouse management module of the target warehouse to ensure smooth data interaction and acquisition, and the relevant information of the pool of goods to be sorted is obtained through communication with the warehouse management module. The pool of goods to be sorted refers to a collection of goods that need to be sorted, and these goods have different characteristics and attributes. At the same time, a list of goods to be stored will be obtained, which contains detailed information about the goods that need to be stored in the warehouse, such as the type, quantity, and specifications of the goods. After obtaining this information, it is sorted and analyzed for subsequent processing and operation. For example, the goods in the pool of goods to be sorted are classified and marked to better carry out subsequent sorting work. It provides important basic data for the cargo identification and sorting method of the entire intelligent warehousing, ensuring that the subsequent steps can be carried out accurately and efficiently.

[0017] Step S200: performing feature recognition on the pool of goods to be sorted according to the Hu-Rajan-Chia model to determine first goods feature information corresponding to the first goods to be sorted.

[0018] Specifically, first, traverse the pool of goods to be sorted, extract the first goods to be sorted from it, use the industrial camera to capture the image of the first goods to be sorted, and obtain its goods image. Next, in order to enhance the quality and characteristics of the goods image, the first goods image is enhanced according to the gamma enhancement function. The gamma enhancement function adjusts the gray value of each pixel in the image to make the contrast and details of the image more obvious. After the enhancement process, the first goods enhanced image is obtained. Finally, based on the first goods enhanced image, the Hu-Rajan-Chia model is used to analyze and identify the features in the image. The model can extract various feature information of the goods, such as shape, color, texture, etc. By identifying and analyzing these features, the first goods feature information corresponding to the first goods to be sorted is determined, and the feature information of the first goods to be sorted is accurately obtained, which provides an important basis for subsequent warehousing authentication and sorting control decisions.

[0019] Step S300: Based on the first cargo characteristic information and according to the list of cargo to be stored, the first cargo to be sorted is authenticated for entry into the warehouse to obtain a first cargo entry authentication result.

[0020] Specifically, based on the acquired characteristic information of the first goods, the first goods to be sorted are authenticated for entry into the warehouse according to the list of goods to be stored. The list of goods to be stored is read to obtain characteristic information of multiple goods to be stored, and then the characteristic information of the first goods and the characteristic information of the goods to be stored are evaluated for similarity. By comparing and analyzing the similarity between the two, multiple first goods storage registration coefficients are obtained. These coefficients reflect the degree of matching between the first goods to be sorted and the goods in the list of goods to be stored. The maximum value of these first goods storage registration coefficients is screened to obtain the first goods entry authentication coefficient, which is an important basis for judging whether the first goods to be sorted can be entered into the warehouse. Subsequently, it is judged whether the first goods entry authentication coefficient is greater than or equal to the goods entry authentication threshold. If yes, then the first cargo warehousing certification result is passed, indicating that the cargo meets the warehousing requirements and subsequent warehousing operations can be carried out; if no, then the first cargo warehousing certification result is failed, and a first cargo warehousing warning instruction will be generated to remind relevant personnel to pay attention to possible problems with the cargo. By comparing and analyzing the characteristic information of the first cargo and the list of cargo to be stored, the warehousing certification of the first cargo to be sorted is achieved, ensuring the accuracy and compliance of the incoming cargo.

[0021] Step S400: When the first cargo warehousing authentication result is passed, the first cargo real-time position and the first cargo storage position corresponding to the first cargo to be sorted are read, and control parameter decisions are made on the sorting robot in combination with the first cargo feature information to obtain a sorting control decision space.

[0022] Specifically, when the result of the first cargo entry authentication is passed, a series of operations are performed to determine the sorting control decision space. The real-time position of the first cargo and the storage position of the first cargo corresponding to the first cargo to be sorted are read. These position information are crucial for the subsequent sorting operation, and the current position of the cargo and the storage position to be delivered are determined. Then, the control parameter decision of the sorting robot is made in combination with the first cargo feature information. The first cargo feature information includes the shape, size, weight, etc. of the cargo. This information will affect the operation mode and path selection of the sorting robot. According to these feature information, the real-time position and storage position of the cargo are comprehensively considered to determine the control parameters of the sorting robot. In the process of making control parameter decisions, multiple sorting control decisions are generated. These decisions represent different sorting schemes and path selections. They constitute the sorting control decision space. These decisions are evaluated and compared to select the optimal sorting control strategy. By reading the cargo location information and making control parameter decisions in combination with the cargo feature information, a space containing multiple sorting control decisions is obtained, which improves the efficiency and accuracy of sorting and ensures that the cargo can be accurately delivered to the designated storage location.

[0023] Step S500: Based on the first cargo sorting loss threshold corresponding to the first cargo to be sorted and the expected sorting efficiency of the first cargo, a multi-level joint optimization is performed on the sorting control decision space to determine the first cargo sorting control strategy.

[0024] Specifically, based on the first cargo sorting loss threshold, the sorting control decision space is optimized for sorting loss, and the first level of sorting control optimization is established. According to the sorting control decision space, the first sorting control decision is extracted, and based on the decision, the sorting loss of the first cargo to be sorted is predicted to obtain the first sorting loss coefficient. It is judged whether the first sorting loss coefficient is less than or equal to the first cargo sorting loss threshold. If so, the first sorting control decision is added to the first level of sorting control optimization; otherwise, the decision is eliminated. According to the first cargo sorting loss threshold, the sorting control decision space is continued to be optimized and analyzed to generate the first level of sorting control optimization. Based on the expected sorting efficiency of the first cargo, the sorting control decision space is optimized for sorting efficiency, and the second level of sorting control optimization is established. According to the sorting control decision space, the nth sorting control decision is extracted. Then, based on the decision, the sorting efficiency of the first cargo to be sorted is predicted to obtain the nth sorting efficiency coefficient, and it is judged whether the nth sorting efficiency coefficient is greater than or equal to the expected sorting efficiency of the first cargo. If so, add the nth sorting control decision to the second level of sorting control optimization; otherwise, eliminate the decision. According to the expected sorting efficiency of the first cargo, continue to perform optimization analysis on the sorting control decision space to generate the second level of sorting control optimization. Then, obtain the intersection of the first level of sorting control optimization and the second level of sorting control optimization to generate the third level of sorting control optimization. Finally, construct a cargo sorting control fitness function, and based on this function, perform sorting control fitness optimization on the third level of sorting control optimization to generate the first cargo sorting control strategy. Through the above multi-level joint optimization process, it is possible to comprehensively consider the sorting loss and sorting efficiency, determine the optimal first cargo sorting control strategy, and thus improve the cargo sorting efficiency and quality of intelligent warehousing.

[0025] Step S600: Based on the sorting robot, the first goods to be sorted are sorted according to the first goods sorting control strategy.

[0026] Specifically, the sorting robot is started and ensured to be in normal working state. The sorting robot will perform the sorting task according to the first cargo sorting control strategy determined previously. According to the first cargo sorting control strategy, the sorting robot will obtain relevant information of the first cargo to be sorted, including the location, characteristics and target location of the cargo. Then, the sorting robot will plan the best sorting path and action sequence based on this information. It will use its own mechanical arm or other grasping device to accurately grasp the first cargo to be sorted. After grasping the cargo, the sorting robot will transport the cargo to the designated target location according to the planned path. During this process, the sorting robot will constantly adjust its actions to ensure that the cargo can be delivered to the destination safely and accurately. After completing the sorting task, the sorting robot will feedback the sorting results, including whether the sorting is successful, the sorting time and other information. Through the precise operation of the sorting robot and the guidance of the first cargo sorting control strategy, the sorting work of the first cargo to be sorted can be completed efficiently and accurately, and the logistics efficiency of intelligent warehousing can be improved.

[0027] In a possible implementation, step S200 further includes:

[0028] Step S210: traverse the pool of goods to be sorted and extract the first goods to be sorted.

[0029] Step S220: collecting a first cargo image of the first cargo to be sorted using an industrial camera.

[0030] Step S230: performing enhancement processing on the first cargo image according to a gamma enhancement function to obtain a first cargo enhanced image.

[0031] Step S240: Based on the first cargo enhanced image and the Hu-Rajan-Chia model, the first cargo feature information is obtained.

[0032] Specifically, the goods in the pool of goods to be sorted are checked one by one, starting from the beginning of the pool of goods to be sorted, and each of the goods is visited in turn. When visiting each goods, it is determined whether the goods are the first goods to be sorted. If so, it is extracted for subsequent processing. This process will continue until all the goods in the pool of goods to be sorted are traversed. In this way, the first goods to be sorted are accurately found and extracted.

[0033] The industrial camera is used to collect images of the first goods to be sorted. The industrial camera is placed in a suitable position to ensure that the first goods to be sorted can be clearly photographed. When the preparations are ready, the industrial camera will start the shooting function and take pictures of the first goods to be sorted. During the shooting process, the industrial camera will quickly capture the appearance characteristics of the goods, including shape, color, texture and other information. Through precise optical systems and sensors, the industrial camera can convert this information into digital image signals.

[0034] In order to improve the quality and clarity of the first cargo image, a gamma enhancement function is used to enhance it, and the grayscale value of each pixel of the first cargo image is substituted into the gamma enhancement function for calculation. The gamma enhancement function adjusts the grayscale value according to preset parameters to enhance the contrast and brightness of the image. In this way, the details in the image are more obvious and the edges are clearer, which is helpful for the subsequent accurate identification of the cargo features. After being processed by the gamma enhancement function, the first cargo enhanced image is obtained, which has better visual effects and information readability than the original image.

[0035] Based on the obtained enhanced image of the first cargo, the Hu-Rajan-Chia model is used to extract the characteristic information of the first cargo. The enhanced image of the first cargo is input into the Hu-Rajan-Chia model, which performs an in-depth analysis of the image, identifies various features and patterns in the image, and considers information in multiple aspects such as the color, shape, and texture of the image. Through comprehensive analysis of this information, the model can extract features related to the first cargo, such as the shape characteristics, color distribution, and surface texture of the cargo. These characteristic information will be acquired and recorded by the system as the characteristic information of the first cargo. These characteristic information are crucial for subsequent operations such as cargo warehousing authentication and sorting control, helping to more accurately identify and process cargo and improve the efficiency and accuracy of warehousing and sorting.

[0036] In a possible implementation, step S200 further includes:

[0037] ;

[0038] in, Characterizing the First Cargo Enhanced Image The gray value at the position, round represents the rounding process, Characterizing the first cargo image The gray value at the position, mean represents the average gray value of the first cargo image, Characterizes the image enhancement gamma coefficient.

[0039] Specifically, It represents the coordinates of the first cargo enhanced image. The grayscale value at the position, round represents the rounding operation, which rounds the calculation result. It represents the same coordinates in the first cargo image The gray value at the position, mean represents the average gray value of the first cargo image, which is obtained by statistically calculating the gray values ​​of all pixels in the image. γ is the image enhancement gamma coefficient, which plays an important role in image enhancement processing. The value of changes the contrast and brightness of the image, thereby achieving the effect of enhancing the image. By processing the gray value of the first cargo image, combining the average gray value and the gamma coefficient, the gray value of a specific position in the first cargo enhanced image is obtained, and the final result is obtained by rounding.

[0040] In one possible implementation, Figure 2 As shown, step S300 also includes:

[0041] Step S310: Read the list of goods to be stored to obtain characteristic information of multiple goods to be stored.

[0042] Step S320: Based on the first cargo feature information, similarity feature evaluation is performed on the plurality of cargo feature information to be stored respectively to obtain a plurality of first cargo storage registration coefficients.

[0043] Step S330: Screening the maximum value according to the plurality of first cargo storage registration coefficients to obtain the first cargo warehousing certification coefficient.

[0044] Step S340: determining whether the first goods entry authentication coefficient is greater than / equal to a goods entry authentication threshold.

[0045] Step S350: If the first cargo entry authentication coefficient is greater than / equal to the cargo entry authentication threshold, the first cargo entry authentication result obtained is passed.

[0046] Step S360: If the first cargo entry authentication coefficient is less than the cargo entry authentication threshold, the first cargo entry authentication result obtained is failure, and a first cargo entry warning instruction is generated.

[0047] Specifically, the reading operation is started, the list of goods to be stored is opened, and the list contains detailed information about the goods. The data in the list is scanned line by line. For each line of data, the content is parsed to extract information related to the characteristics of the goods, including the name, model, specification, quantity, production date, shelf life, etc. of the goods. By reading and parsing the list, the characteristic information of multiple goods to be stored is obtained, and the information is stored in the system's memory or database for processing and analysis in subsequent steps.

[0048] It is clear that the characteristic information of the first goods can be expressed as a set of specific values, including characteristic information of multiple dimensions such as shape characteristic values, color characteristic values, and size characteristic values. For each of the goods to be stored, there is also a similar set of characteristic information. Then, the Euclidean distance between the first goods and each of the goods to be stored is calculated. For a product to be stored, the square of the difference between it and the first product in each characteristic dimension is calculated respectively, and these square values ​​are added, and then the square root of the sum is taken to obtain the Euclidean distance between the two. For example, the shape feature is represented by a digital code, the color feature is represented by a specific color value, and the size feature is represented by the length, width, and height values. Then, for the shape feature dimension, the square of the difference between the shape codes of the two is calculated; for the color feature dimension, the square of the difference in color values ​​is calculated; for the size feature dimension, the square of the size difference in the three directions of length, width, and height is calculated respectively, and then all these square values ​​are added together, and finally the square root is taken to obtain the Euclidean distance. The smaller the Euclidean distance is, the smaller the difference between the two goods in each characteristic dimension is, that is, the higher the degree of similarity is. According to the calculated Euclidean distance, the first cargo storage registration coefficient is determined, and the Euclidean distance is converted so that the smaller the distance, the larger the registration coefficient. For example, the registration coefficient is calculated using the formula registration coefficient = 1 / (1+Euclidean distance). In this way, when the Euclidean distance is 0, the registration coefficient is 1, indicating complete similarity; as the Euclidean distance increases, the registration coefficient gradually decreases. By performing such calculations for each cargo to be stored, multiple first cargo storage registration coefficients are obtained. These coefficients reflect the similarity between the first cargo and each cargo to be stored, providing a basis for subsequent operations such as warehousing authentication.

[0049] The multiple first goods storage registration coefficients are analyzed and processed, and the values ​​of these registration coefficients are checked one by one. These coefficients reflect the degree of matching between the first goods and the characteristics of different goods to be stored, and the goal is to find the largest registration coefficient. By comparing the size of each coefficient, the maximum value is determined. This maximum value is considered to be the coefficient corresponding to the most matching situation between the first goods and all the goods to be stored, and this maximum value is defined as the first goods storage certification coefficient.

[0050] The first goods warehousing certification coefficient is compared with the goods warehousing certification threshold. This certification threshold is a standard value set according to the requirements and experience of warehouse management, and is used to determine whether the goods can meet the warehousing conditions.

[0051] The judgment result is that the first cargo warehousing certification coefficient is greater than or equal to the cargo warehousing certification threshold, indicating that after the first cargo is evaluated for similar characteristics with the cargo in the list of cargo to be stored, its matching degree meets the requirements for warehousing. At this time, the first cargo warehousing certification result is passed, which means that the first cargo can proceed with the subsequent warehousing operation process, such as determining the storage location, arranging transportation, etc.

[0052] If the first cargo entry certification coefficient is less than the cargo entry certification threshold, it means that the first cargo is significantly different from the cargo in the list of cargo to be stored, or does not meet the specific standards for entry. In this case, the first cargo entry certification result is failure, and a warning instruction for the first cargo entry is generated. This warning instruction reminds warehouse management personnel to conduct further inspection and processing of the cargo, and it is necessary to re-evaluate the characteristics and source of the cargo or communicate and coordinate with relevant parties to determine whether other measures can be taken to make the cargo meet the entry conditions, or to handle the cargo in other ways, such as returning or temporarily storing it for further processing.

[0053] In a possible implementation, step S500 further includes:

[0054] Step S510: Based on the first cargo sorting loss threshold, the sorting loss optimization is performed on the sorting control decision space to establish a first level of sorting control optimization.

[0055] Step S520: Based on the expected sorting efficiency of the first cargo, the sorting efficiency is optimized in the sorting control decision space, and a second layer of sorting control optimization is established.

[0056] Step S530: Obtain the intersection of the first level of sorting control optimization and the second level of sorting control optimization to generate a third level of sorting control optimization.

[0057] Step S540: constructing a cargo sorting control fitness function.

[0058] Step S550: performing sorting control fitness optimization on the third level of sorting control optimization based on the cargo sorting control fitness function to generate the first cargo sorting control strategy.

[0059] Specifically, sorting records are retrieved based on the sorting robot to obtain the sorting cargo feature information record set, sorting control decision record set and sorting loss coefficient record set. These record sets contain various data of past sorting operations and are the basis for training deep neural networks. The sorting cargo feature information record set and the sorting control decision record set are used as input information. They include the size, weight, shape and other characteristics of the cargo and the corresponding sorting control decisions, such as the robot's grasping force and speed. The sorting loss coefficient record set is used as output information, representing the actual degree of sorting loss under specific cargo characteristics and control decisions. The deep neural network is supervised and trained, and the sorting loss prediction error is obtained every predetermined number of trainings. By continuously adjusting the weights and parameters of the network, the prediction error is gradually reduced. When the sorting loss prediction error is less than the sorting loss error threshold, it indicates that the network has learned a sufficiently good pattern and generated a sorting loss prediction model. When predicting the sorting loss of the first cargo to be sorted, based on the first cargo feature information and the first sorting control decision currently being considered, it is input into the trained deep neural network, and the network outputs a predicted sorting loss coefficient, which reflects the degree of sorting loss that may be incurred by the first cargo under the control decision. If the sorting loss coefficient is less than or equal to the sorting loss threshold of the first cargo, the sorting control decision is added to the first layer of sorting control optimization, providing feasible decision options for the subsequent optimization process.

[0060] Collect various data related to the first cargo to be sorted, including historical sorting efficiency data, cargo feature information, environmental factors, etc. under different sorting control decisions. These data will serve as the basis for training the linear regression model. Quantify the nth sorting control decision in the sorting control decision space, for example, convert the robot's speed setting, gripping force and other parameters into numerical features, and extract the cargo feature information corresponding to the sorting control decision, such as the weight, volume, shape, etc. of the cargo. Take these quantized sorting control decision features and cargo features as input variables, and take the sorting efficiency in the historical data as the output variable to train the linear regression model. By adjusting the parameters of the model, the model can predict the sorting efficiency under specific input as accurately as possible. When predicting the sorting efficiency of the first cargo to be sorted, input the current nth sorting control decision and the feature information of the first cargo to be sorted into the trained linear regression model, and the model will output a predicted nth sorting efficiency coefficient, which reflects the sorting efficiency level that the first cargo to be sorted may achieve under the sorting control decision. If this sorting efficiency coefficient is greater than or equal to the expected sorting efficiency of the first cargo, the sorting control decision is added to the second level of sorting control optimization; if it is less than the expected sorting efficiency of the first cargo, the decision is eliminated.

[0061] After the previous steps, the first level of sorting control optimization is obtained by optimizing based on the sorting loss threshold of the first cargo, which contains a set of sorting control decisions that meet the sorting loss requirements. These decisions can ensure that the sorting loss is controlled within an acceptable range when the first cargo to be sorted is sorted. At the same time, the second level of sorting control optimization is obtained by optimizing based on the expected sorting efficiency of the first cargo, in which the decision set can meet the target requirements of sorting efficiency and ensure that the sorting task of the first cargo to be sorted can be completed efficiently within a unit time. Then, the intersection of these two levels is obtained, and only the sorting control decisions that meet both the sorting loss requirements and the sorting efficiency requirements will be retained and enter the third level of sorting control optimization. In this way, the range of feasible sorting control decisions is further narrowed, and the possibility of finding the optimal sorting control strategy is increased. Compare the sorting control decisions in the two levels one by one. If a decision exists in both the first and second levels, then this decision will be added to the third level of sorting control optimization. The third-level decision set obtained in this way not only takes into account the limitation of sorting loss, but also meets the goal of sorting efficiency, providing a more accurate basis for the subsequent construction of cargo sorting control fitness function and fitness optimization.

[0062] The purpose of constructing the cargo sorting control fitness function is to comprehensively evaluate the pros and cons of different sorting control decisions. This fitness function usually takes into account multiple key factors, the most important of which are sorting efficiency and sorting loss.

[0063] The sorting efficiency coefficient reflects how fast the first cargo can be sorted under a specific sorting control decision. High sorting efficiency means that more cargo can be processed in a unit time, which is crucial to improving the efficiency of warehouse operations. The sorting loss coefficient measures the damage, loss or wrong sorting of cargo that may occur during the sorting process. A low sorting loss coefficient indicates that the sorting process is more reliable and can reduce the loss and waste of cargo. The weight conditions α and β in the fitness function are used to adjust the relative importance of sorting efficiency and sorting loss in the overall evaluation. These weights can be set flexibly according to actual warehousing needs and business goals. If warehouse management pays more attention to fast sorting, the value of α will be increased to increase the proportion of sorting efficiency in fitness evaluation; if the integrity requirements of the cargo are high, the value of β will be increased, and more attention will be paid to the control of sorting loss. The introduction of the exponential function (exp) with the natural constant e as the base makes the fitness function have certain nonlinear characteristics, which better reflects the complex relationship between different factors. By properly normalizing the sorting efficiency and sorting loss coefficient and substituting them into the fitness function, a value can be obtained, which represents the fitness of a specific sorting control decision. The higher the fitness value, the better the decision performs in terms of comprehensively considering the sorting efficiency and sorting loss.

[0064] Based on the constructed cargo sorting control fitness function, the third layer of sorting control optimization is optimized for sorting control fitness. The third layer of sorting control optimization is obtained by finding the intersection of the first and second layers of sorting control optimization, in which the sorting control decisions meet both the sorting loss requirements and the sorting efficiency requirements. For each sorting control decision in the third layer of sorting control optimization, substitute it into the cargo sorting control fitness function and calculate the corresponding fitness value. Then, by comparing these fitness values, the decision with the highest fitness is selected as the first cargo sorting control strategy. This strategy will provide the best control parameters for the sorting robot on the basis of comprehensive consideration of sorting efficiency and sorting loss, so as to achieve efficient and low-loss cargo sorting operations. For example, assuming that there are multiple sorting control decisions in the third layer of sorting control optimization, after calculating their fitness values ​​respectively, it is found that the fitness value of a decision is significantly higher than that of other decisions. This decision has a better combination in terms of speed setting, grasping force, path planning, etc. of the sorting robot, and can achieve higher sorting efficiency while ensuring lower sorting loss. Therefore, this decision is selected as the first cargo sorting control strategy to guide the sorting robot to perform sorting operations on the first cargo to be sorted.

[0065] In a possible implementation, step S510 further includes:

[0066] Step S511: extracting a first sorting control decision according to the sorting control decision space.

[0067] Step S512: Based on the first sorting control decision, a sorting loss prediction is performed on the first cargo to be sorted to obtain a first sorting loss coefficient.

[0068] Step S513: Determine whether the first sorting loss coefficient is less than / equal to the first cargo sorting loss threshold.

[0069] Step S514: if the first sorting loss coefficient is less than / equal to the first cargo sorting loss threshold, the first sorting control decision is added to the first level of the sorting control optimization.

[0070] Step S515: If the first sorting loss coefficient is less than / equal to the first cargo sorting loss threshold, the first sorting control decision is eliminated.

[0071] Step S516: According to the first cargo sorting loss threshold, continue to perform optimization analysis on the sorting control decision space to generate the first level of sorting control optimization.

[0072] Specifically, a first sorting control decision is extracted from the sorting control decision space. This decision space contains various possible sorting control schemes, such as different sorting robot speed settings, gripping forces, path planning, etc. The first sorting control decision is a randomly selected decision.

[0073] Based on this first sorting control decision, a sorting loss prediction is performed for the first cargo to be sorted. This involves using a deep neural network model to estimate the possible loss when sorting under this decision, taking into account factors such as cargo characteristics, sorting environment, and parameters of the sorting control decision. Through this prediction, a first sorting loss coefficient is obtained, which quantifies the degree of sorting loss.

[0074] Then, it is determined whether the first sorting loss coefficient is less than or equal to the first cargo sorting loss threshold. This threshold is set according to warehouse management requirements and experience, and represents the maximum acceptable sorting loss level.

[0075] If the judgment result is that the first sorting loss coefficient is less than or equal to the first cargo sorting loss threshold, then the first sorting control decision is added to the first level of sorting control optimization. This means that the decision is more feasible in terms of sorting loss and may become part of the final sorting control strategy.

[0076] If the judgment result is that the first sorting loss coefficient is greater than the first cargo sorting loss threshold, then the first sorting control decision is eliminated because it cannot meet the sorting loss requirement.

[0077] According to the first cargo sorting loss threshold, the sorting control decision space is continuously optimized, which means that new sorting control decisions are continuously extracted from the decision space, and the above steps are repeated until the entire decision space is traversed. The first layer of sorting control optimization finally generated will contain those sorting control decisions that perform well in terms of sorting loss.

[0078] In a possible implementation, step S512 further includes:

[0079] Step S5121: performing sorting record retrieval based on the sorting robot to obtain a sorting cargo feature information record set, a sorting control decision record set and a sorting loss coefficient record set.

[0080] Step S5122: Using the sorting cargo feature information record set and the sorting control decision record set as input information and the sorting loss coefficient record set as output information, supervised training is performed on the deep neural network, and a sorting loss prediction error is obtained after each predetermined number of trainings.

[0081] Step S5123: When the sorting loss prediction error is less than the sorting loss error threshold, a sorting loss prediction model is generated.

[0082] Step S5124: Based on the first cargo feature information and the first sorting control decision, the first sorting loss coefficient is output according to the sorting loss prediction model.

[0083] Specifically, sorting records are retrieved based on the sorting robot to query the records of goods handled by the sorting robot in the past sorting operations. Through the retrieval, a set of sorted goods feature information records is obtained, which contains the feature descriptions of various sorted goods, such as the size, weight, shape, material and other information of the goods. At the same time, a set of sorting control decision records can also be obtained, which records the specific control decisions adopted when sorting different goods, such as the robot's grasping force, moving speed, path planning, etc. In addition, a set of sorting loss coefficient records is obtained, which records the degree of loss corresponding to each sorting operation, including indicators such as the damage of the goods and the number of incorrect sorting.

[0084] First, it is clear that the input information is the sorting cargo feature information record set and the sorting control decision record set, and the output information is the sorting loss coefficient record set. The deep neural network is supervised and trained using these data to enable the neural network to learn the complex mapping relationship between cargo feature information, sorting control decisions and sorting loss coefficients. The sorting cargo feature information record set contains detailed feature descriptions of various sorted cargoes, such as cargo size, weight, shape, material and other attributes. The sorting control decision record set records the specific sorting control decisions adopted for different cargoes, such as the robot's grasping force, moving speed, path planning and other parameter settings. By providing these rich input information to the neural network, it can understand the factors affecting sorting loss from different angles. The sorting loss coefficient record set, as output information, represents the actual degree of sorting loss under specific cargo features and sorting control decisions. During the training process, the neural network continuously adjusts its weights and parameters to predict the sorting loss coefficient corresponding to the given input information as accurately as possible. After each predetermined number of training, the sorting loss prediction error is calculated. This error is obtained by comparing the sorting loss coefficient currently predicted by the neural network with the sorting loss coefficient actually recorded. It reflects the degree of fit of the neural network to the data in the current training stage. If the predicted value differs greatly from the actual value, it means that the error is large and the network needs further adjustment to improve the prediction accuracy; if the error is small, it means that the network is gradually learning the correct mapping relationship. By continuously monitoring and reducing this error, the neural network can continuously optimize its own performance and better adapt to different cargo characteristics and sorting control decision combinations, thus laying the foundation for accurately predicting the sorting loss coefficient of the first cargo to be sorted.

[0085] When the sorting loss prediction error is less than the sorting loss error threshold, it indicates that the deep neural network has learned a good enough pattern to generate a sorting loss prediction model. This model can accurately predict the sorting loss coefficient based on the input cargo feature information and sorting control decisions.

[0086] Based on the first cargo feature information and the first sorting control decision, this information is input into the generated sorting loss prediction model. The model will output the first sorting loss coefficient for the first cargo to be sorted under the specific first sorting control decision according to the learned relationship. This coefficient will be used to judge whether the sorting control decision meets the sorting loss requirement in the future, so as to decide whether to add it to the first level of sorting control optimization.

[0087] In a possible implementation, step S520 further includes:

[0088] Step S521: extracting the nth sorting control decision according to the sorting control decision space, where n is a positive integer.

[0089] Step S522: Based on the nth sorting control decision, a sorting efficiency prediction is performed on the first cargo to be sorted to obtain an nth sorting efficiency coefficient.

[0090] Step S523: Determine whether the nth sorting efficiency coefficient is greater than / equal to the first cargo expected sorting efficiency.

[0091] Step S524: If the nth sorting efficiency coefficient is greater than / equal to the first cargo expected sorting efficiency, the nth sorting control decision is added to the second level of the sorting control optimization.

[0092] Step S525: If the nth sorting efficiency coefficient is less than the first cargo expected sorting efficiency, the nth sorting control decision is eliminated.

[0093] Step S526: According to the expected sorting efficiency of the first cargo, continue to perform optimization analysis on the sorting control decision space to generate the second level of sorting control optimization.

[0094] Specifically, the nth sorting control decision is extracted from the sorting control decision space, where n is a positive integer, indicating that multiple decisions in the sorting control decision space are analyzed in sequence. The sorting control decision space contains various possible combinations of sorting control solutions, such as different combinations of robot movement speeds, grasping methods, path planning and other parameters.

[0095] Based on the extracted nth sorting control decision, the sorting efficiency of the first to-be-sorted goods is predicted, and the linear regression model is used to estimate the efficiency that can be achieved when sorting the first to-be-sorted goods under this decision, taking into account factors such as the characteristics of the goods, the sorting environment, and the parameters of the specific sorting control decision. An nth sorting efficiency coefficient is obtained through prediction, which quantifies the sorting efficiency level under this sorting control decision.

[0096] It is determined whether the nth sorting efficiency coefficient is greater than or equal to the first cargo expected sorting efficiency. The first cargo expected sorting efficiency is a target value set according to the warehousing business requirements and actual conditions, representing the sorting speed and efficiency level expected to be achieved for the first cargo to be sorted.

[0097] If the nth sorting efficiency coefficient is greater than or equal to the expected sorting efficiency of the first cargo, it means that the sorting control decision is feasible in terms of sorting efficiency and can meet or exceed the expected target. At this time, the nth sorting control decision is added to the second level of sorting control optimization. The second level of sorting control optimization is a set of decisions specially established to meet the sorting efficiency requirements.

[0098] If the nth sorting efficiency coefficient is less than the expected sorting efficiency of the first cargo, it means that the decision cannot meet the sorting efficiency requirements. In this case, the nth sorting control decision is eliminated and will not be included in the subsequent consideration scope.

[0099] According to the expected sorting efficiency of the first cargo, the sorting control decision space is continuously optimized. This means that new sorting control decisions are continuously extracted from the sorting control decision space, and the process from step S521 to S525 is repeated until enough decisions that meet the sorting efficiency requirements are found, or the entire sorting control decision space is traversed. The second level of sorting control optimization finally generated will contain those sorting control decisions that perform well in terms of sorting efficiency.

[0100] In a possible implementation, step S540 further includes:

[0101] ;

[0102] Among them, SGF represents the fitness of cargo sorting control, exp represents the exponential function with the natural constant e as the base, α and β represent the weight conditions for calculating the fitness of sorting control, K (sge) represents the normalized sorting efficiency coefficient, and K (sgl) represents the normalized sorting loss coefficient.

[0103] Specifically, SGF is used to measure the overall performance of a specific cargo sorting control strategy. The higher the fitness, the better the sorting control strategy can achieve a balance between sorting efficiency and sorting loss, and thus it is more suitable for actual cargo sorting operations. exp represents an exponential function with the natural constant e as the base, which performs a nonlinear transformation on the input value, thereby more sensitively reflecting the relationship between different factors. α and β represent the weight conditions for sorting control fitness calculation, which are adjusted according to actual warehousing needs and business goals. If you pay more attention to sorting efficiency, increase the value of α to make the sorting efficiency coefficient occupy a greater weight in the fitness calculation; if you pay more attention to the control of sorting loss, increase the value of β to increase the importance of the sorting loss coefficient in the fitness calculation. K (sge) represents the normalized sorting efficiency coefficient. The normalization process converts the sorting efficiency coefficients of different dimensions into values ​​within a specific range, which is convenient for comprehensive calculation with other factors. The higher the sorting efficiency coefficient, the faster the goods can be sorted under the sorting control strategy, thereby improving the operational efficiency of warehousing. K(sgl) represents the normalized sorting loss coefficient. After normalization, the sorting loss coefficient can more accurately reflect the degree of loss that goods may suffer under a specific sorting control strategy. The lower the sorting loss coefficient, the less likely the goods are to be damaged, lost or missorted during the sorting process.

[0104] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0105] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0106] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for identifying and sorting goods in intelligent warehousing, characterized in that: The method comprises: Interact with the storage management module of the target warehouse to obtain the pool of goods to be sorted and the list of goods to be stored; Performing feature recognition on the pool of goods to be sorted according to the Hu-Rajan-Chia model to determine first goods feature information corresponding to the first goods to be sorted; Based on the first cargo characteristic information, performing warehousing authentication on the first cargo to be sorted according to the list of cargo to be stored, and obtaining a first cargo warehousing authentication result; When the first cargo entry authentication result is passed, the first cargo real-time position and the first cargo storage position corresponding to the first cargo to be sorted are read, and the control parameter decision of the sorting robot is made in combination with the first cargo feature information to obtain the sorting control decision space; Based on the first cargo sorting loss threshold corresponding to the first cargo to be sorted and the expected sorting efficiency of the first cargo, a multi-level joint optimization is performed on the sorting control decision space to determine the first cargo sorting control strategy; Based on the sorting robot, sorting the first goods to be sorted according to the first goods sorting control strategy; Among them, based on the first cargo sorting loss threshold corresponding to the first cargo to be sorted and the expected sorting efficiency of the first cargo, a multi-level joint optimization is performed on the sorting control decision space to determine the first cargo sorting control strategy, including: Based on the first cargo sorting loss threshold, performing sorting loss optimization on the sorting control decision space to establish a first level of sorting control optimization; Based on the expected sorting efficiency of the first cargo, optimizing the sorting efficiency in the sorting control decision space, and establishing a second layer of sorting control optimization; Obtaining the intersection of the first level of sorting control optimization and the second level of sorting control optimization to generate a third level of sorting control optimization; Construct cargo sorting control fitness function; The third level of sorting control optimization is optimized for sorting control fitness based on the cargo sorting control fitness function to generate the first cargo sorting control strategy.

2. The method according to claim 1, characterized in that Performing feature recognition on the pool of goods to be sorted according to the Hu-Rajan-Chia model to determine first goods feature information corresponding to the first goods to be sorted includes: Traversing the pool of goods to be sorted, and extracting the first goods to be sorted; Collecting a first cargo image of the first cargo to be sorted using an industrial camera; Performing enhancement processing on the first cargo image according to a gamma enhancement function to obtain a first cargo enhanced image; Based on the first cargo enhanced image and according to the Hu-Rajan-Chia model, the first cargo feature information is obtained.

3. The method according to claim 2, characterized in that The gamma enhancement function is: Among them, g(x,y) represents the grayscale value at the position (x,y) in the first cargo enhanced image, round represents the rounding process, f(x,y) represents the grayscale value at the position (x,y) in the first cargo image, mean represents the average grayscale value of the first cargo image, and γ represents the image enhancement gamma coefficient.

4. The method according to claim 1, characterized in that Based on the first cargo characteristic information, performing warehousing authentication on the first cargo to be sorted according to the list of cargo to be stored, and obtaining a first cargo warehousing authentication result, including: Read the list of goods to be stored to obtain characteristic information of multiple goods to be stored; Based on the first cargo feature information, respectively perform similarity feature evaluation on the plurality of cargo feature information to be stored, and obtain a plurality of first cargo storage registration coefficients; Screening the maximum value according to the plurality of first cargo storage registration coefficients to obtain the first cargo storage certification coefficient; Determining whether the first cargo entry authentication coefficient is greater than / equal to a cargo entry authentication threshold; If the first cargo warehousing authentication coefficient is greater than / equal to the cargo warehousing authentication threshold, the first cargo warehousing authentication result obtained is passed; If the first cargo entry authentication coefficient is less than the cargo entry authentication threshold, the first cargo entry authentication result obtained is failure, and a first cargo entry warning instruction is generated.

5. The method according to claim 1, characterized in that Based on the first cargo sorting loss threshold, the sorting control decision space is optimized for sorting loss, and a first level of sorting control optimization is established, including: Extracting a first sorting control decision according to the sorting control decision space; Based on the first sorting control decision, a sorting loss prediction is performed on the first goods to be sorted to obtain a first sorting loss coefficient; Determining whether the first sorting loss coefficient is less than / equal to the first cargo sorting loss threshold; If the first sorting loss coefficient is less than / equal to the first cargo sorting loss threshold, adding the first sorting control decision to the first level of sorting control optimization; If the first sorting loss coefficient is less than / equal to the first cargo sorting loss threshold, the first sorting control decision is eliminated; According to the first cargo sorting loss threshold, the sorting control decision space is continuously optimized and analyzed to generate the first level of sorting control optimization.

6. The method according to claim 5, characterized in that Based on the first sorting control decision, a sorting loss prediction is performed on the first to-be-sorted goods to obtain a first sorting loss coefficient, including: Performing sorting record retrieval based on the sorting robot to obtain a sorting cargo feature information record set, a sorting control decision record set, and a sorting loss coefficient record set; Taking the sorting cargo feature information record set and the sorting control decision record set as input information and the sorting loss coefficient record set as output information, supervised training is performed on the deep neural network, and a sorting loss prediction error is obtained after each predetermined number of trainings; When the sorting loss prediction error is less than the sorting loss error threshold, generating a sorting loss prediction model; Based on the first cargo feature information and the first sorting control decision, the first sorting loss coefficient is output according to the sorting loss prediction model.

7. The method according to claim 1, characterized in that Based on the expected sorting efficiency of the first cargo, the sorting efficiency is optimized in the sorting control decision space, and a second layer of sorting control optimization is established, including: Extracting the nth sorting control decision according to the sorting control decision space, where n is a positive integer; Based on the nth sorting control decision, predict the sorting efficiency of the first to-be-sorted goods to obtain an nth sorting efficiency coefficient; Determining whether the nth sorting efficiency coefficient is greater than / equal to the first cargo expected sorting efficiency; If the nth sorting efficiency coefficient is greater than / equal to the first cargo expected sorting efficiency, adding the nth sorting control decision to the second level of sorting control optimization; If the nth sorting efficiency coefficient is less than the first cargo expected sorting efficiency, eliminating the nth sorting control decision; According to the expected sorting efficiency of the first cargo, the sorting control decision space is further optimized and analyzed to generate the second level of sorting control optimization.

8. The method according to claim 1, characterized in that The cargo sorting control fitness function is: ； Among them, SGF represents the fitness of cargo sorting control, exp represents the exponential function with the natural constant e as the base, α and β represent the weight conditions for calculating the fitness of sorting control, K (sge) represents the normalized sorting efficiency coefficient, and K (sgl) represents the normalized sorting loss coefficient.

Images