Physical Trusted Traceability Warehouse Management Device Based on Unstructured Blockchain Features

Abstract

The present invention relates to the field of warehousing management, specifically a physical trusted traceability warehousing management device based on unstructured blockchain features, including: realizing the automatic extraction of various physical process feature data for the WMS application scenario through machine vision methods; the secondary fusion process of unstructured data and structured data; constructing a physical trusted distributed inventory system based on blockchain for the WMS application scenario, and giving classification results during the traceability process. The physical trusted traceability warehousing management device based on unstructured blockchain features of the present invention realizes the automatic extraction of various physical process feature data for the WMS application scenario through machine vision methods, realizes the extraction of various physical features, and realizes the physical trusted traceability of physical trusted description, ensuring that each physical process in the WMS system can be uniformly managed and accurately traced, and avoiding the occurrence of physical forgery processes.

Description

Technical Field

[0001] The present invention relates to the field of warehouse management, and specifically to a physical trusted traceability warehouse management device based on unstructured blockchain features. Background Art

[0002] With the gradual implementation of the "Three Types and Two Networks" (hub type, platform type, sharing type; strong smart grid, ubiquitous power Internet of Things) strategic layout of the State Grid, the logistics warehouse management system in the power grid field is also rapidly transforming into intelligent warehouse management systems such as WMS. Through the informatization and intelligent transformation of the original warehouse management system, it has gradually evolved from the automated processes of batch management, material correspondence, inventory counting, and quality inspection management in traditional warehouse management to intelligent capabilities such as virtual warehouse management and real-time inventory management. By effectively controlling and tracking the entire process of logistics and cost management in warehouse operations, it comprehensively serves the strategic layout requirements of the State Grid.

[0003] As a breakthrough technology for distributed traceability, blockchain technology is widely used in various industries. The mathematical principle of blockchain data security is based on the encryption of data using asymmetric cryptography principles, and at the same time, it relies on the powerful computing power formed by consensus algorithms such as the workload proof of each node in the distributed system to resist external attacks.

[0004] The integration of existing blockchain technology and Internet of Things technology is based on the assumption of structured data: that is, Internet of Things technology means extract the structured characteristics of various warehouse objects through technologies such as two-dimensional codes, while blockchain technology simultaneously ensures the trusted traceability process of the above-mentioned structured characteristics. However, such a structured data mapping process often cannot directly correspond to the physical process. That is, when there is physical fraud in the warehouse management process data, such as physical substitution of goods, damage to the original voucher of two-dimensional code structured data, and physical replication of two-dimensional code structured data, the pure logical mapping process cannot achieve true physical trusted traceability. Therefore, for structured data above the logic layer of common cloud platforms, the combination of Internet of Things data and blockchain traceability cannot meet the needs of various physical anti-counterfeiting processes and cannot cope with traceability attacks in various physical fraud processes, resulting in the fact that the traceability process still cannot achieve the effectiveness of the physical process for WMS warehouse management. Therefore, in view of the above situation, there is an urgent need to develop a physical trusted traceability warehouse management device based on unstructured blockchain features to overcome the deficiencies in current practical applications. Summary of the Invention

[0005] The purpose of the present invention is to provide a physical trusted traceability warehouse management device based on unstructured blockchain features to solve the problems raised in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] A physical trusted traceability warehousing management device based on unstructured blockchain features, comprising:

[0008] Through machine vision methods, realize the automatic extraction of various physical process feature data for the WMS application scenario;

[0009] The secondary fusion process of unstructured data and structured data;

[0010] For the WMS application scenario, construct a physical trusted distributed inventory system based on blockchain, and give classification results during the traceability process.

[0011] Compared with the prior art, the beneficial effects of the present invention are:

[0012] 1) Realize the extraction of various physical features;

[0013] 2) Realize the unified secondary mapping of unstructured data and structured data, and realize a unified secondary structured feature data structure, which can efficiently realize the physical trusted description of various process parameters;

[0014] 3) Use logical layer operations such as blockchain token technology to realize the physical trusted traceability of the above physical trusted description, ensure that each physical process in the WMS system can be uniformly managed and accurately traced, and avoid the generation of physical forgery processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is the algorithm flow chart of the RMPE model in the embodiment of the present invention.

[0016] Figure 2 It is the structure schematic diagram of STN SPPE SDTN in the embodiment of the present invention.

[0017] Figure 3 It is the algorithm flow chart of the OpenPose in the embodiment of the present invention.

[0018] Figure 4 It is the framework diagram of the MoveNet prediction line calculation method in the embodiment of the present invention.

[0019] Figure 5 It is the schematic diagram of the alignment process based on Euclidean distance in the embodiment of the present invention.

[0020] Figure 6 It is the schematic diagram of the alignment process of DTW distance in the embodiment of the present invention.

[0021] Figure 7 It is the schematic diagram of the DTW regularization path in the embodiment of the present invention.

[0022] Figure 8 It is the schematic diagram of restricting the search space in the embodiment of the present invention.

[0023] Figure 9 This is a schematic diagram of the data abstraction process in the embodiment of the present invention.

[0024] Figure 10 This is a schematic diagram of the execution flow of the FastDTW algorithm in the embodiment of the present invention.

[0025] Figure 11 This is a schematic diagram of the time axis for aligning sensors by DTW in the embodiment of the present invention. Detailed implementation manners

[0026] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0027] The following describes the specific implementation of the present invention in detail in conjunction with specific embodiments.

[0028] Embodiment 1

[0029] The physical trusted traceability warehousing management device based on unstructured blockchain features provided by the embodiment of the present invention includes:

[0030] Through machine vision methods, realize the automatic extraction of various physical process feature data for the WMS application scenario;

[0031] The secondary fusion process of unstructured data and structured data;

[0032] For the WMS application scenario, construct a physical trusted distributed inventory system based on the blockchain, and give classification results during the traceability process.

[0033] In the process of extracting feature data for WMS application scenarios, the physical layer trusted traceability process is affected not only by Internet of Things structured data (such as two-dimensional codes, etc.), but also by various types of unstructured data that can be obtained from distributed machine vision. Through machine vision methods, the automatic extraction of feature data of various physical processes is realized for WMS application scenarios, the extraction of various physical features (without physical meaning) is achieved, and through the secondary fusion process of unstructured data and structured data, the unified secondary mapping of unstructured data and structured data is realized, and a unified secondary structured feature data structure is achieved, which can efficiently realize the physical trusted description of various process parameters. Through logical layer operations such as using blockchain token technology, the physical trusted traceability of the physical trusted description is realized, ensuring that each physical process in the WMS system can be uniformly managed and accurately traced, avoiding the occurrence of physical forgery processes, and is worthy of promotion.

[0034] Embodiment 2

[0035] Please refer to Figures 1-11 , the automatic extraction process of features for physical processes can be divided into the following three steps:

[0036] Step 1: Through the neural network structure used in machine learning methods such as machine vision, automatically extract features from unstructured data, including the mapping of unstructured data such as the physical behavior of operators, information changes in the physical process of goods transfer, and changes in target shelves.

[0037] Step 2: The above mapping data is regarded as the internal representation parameters of the neural network. Without analyzing its clear physical meaning, it is directly regarded as the result of the extraction of physical process features, and its significance lies in extracting physical process feature data without direct physical meaning.

[0038] Step 3: The extraction results of the above physical processes can be subjected to feature alignment and fusion of distributed multi-nodes under the time and space benchmarks of the WMS system mentioned in the present invention; the fusion process can be simple feature superposition, feature semantic fusion, or feature secondary analysis fusion.

[0039] Collecting unstructured features of physical processes in the traceability process through machine learning methods and completing the logical merging with the traceability process is an effective means to solve the problem that the existing traceability system cannot perform physical authenticity mapping through physical processes. Specifically, it refers to forming a parallel fusion with the traceability process by fusing any unstructured feature data related to the traceability process, such as extracting unstructured data of human behavior, equipment behavior, and logistics transfer in the traceability process, so as to obtain a trusted physical process.

[0040] Among them, taking the recognition of the actor's posture and other characteristics as an example, the actor's posture recognition generally includes two aspects: action recognition and identity recognition. The key to recognition is the extraction of the actor's characteristics. Among them, the detection of the key characteristics of the actor's skeleton is often used as the basic component of posture recognition and is the main aspect of current research.

[0041] The detection of the key characteristics of the actor's skeleton is a multi-faceted task, including object detection, the detection of the key characteristics of the actor's skeleton, segmentation, etc. Among them, the detection of the key characteristics of the actor's skeleton can be divided into 2D and 3D detections; according to the detection method, it can also be divided into two ways: bottom-up (TopDown) and top-down (BottomUp). For the currently more studied 2D detection, its algorithm models basically follow two ideas: top-down and bottom-up to achieve.

[0042] The top-down algorithm for detecting the key characteristics of the actor's skeleton includes two parts: object detection and single-actor key skeleton feature detection. The object detection algorithm outputs the actor's bounding box. The current algorithms such as the Yolo series have high accuracy and speed, and mainly focus on the algorithm for detecting the key characteristics of a single actor's skeleton; through the RMPE model, problems such as possible positioning errors in the bounding box of object detection and repeated detection of the same object are mainly solved. Figure 1 The structures of STN, SPPE, and SDTN in Figure 2 are shown as follows. Through the spatial transformation network, different Proposals generated by the same actor are transformed into a better area to solve the problem of different key feature detection results due to different Proposals generated by an actor. STN extracts a high-quality single-actor candidate box from an inaccurate bounding box, SPPE generates various postures, SDTN generates candidate postures, and at the same time, a parallel ParallelSPPE branch is introduced for optimization.

[0043] The bottom-up algorithm for detecting the key characteristics of the actor's skeleton mainly includes two parts: key feature detection and key feature clustering connection. Key feature detection is to detect all the key features of all people in the picture. After key feature detection, these key features need to be clustered to connect the different key features of each person together, thus connecting to generate different individuals.

[0044] By modeling the different limb structures of the actor and using a vector field to simulate different limb structures, the problem of incorrect connection caused by simply using whether the middle point is on the limb trunk is solved. OpenPose uses this method to detect the key characteristics of the actor's skeleton. This algorithm inputs the image into a two-branch CNN network and jointly outputs the confidence of the joint points and PAF asFigure 3 As shown in Figures (b) and (c) in , the heat map of key features and the associated regions are vectorially connected to form the true skeletal structure of the actor. A spanning tree graph of the individual pose is obtained with a minimum number of edges (using a bipartite graph + Hungarian algorithm, etc.), which greatly reduces the complexity while ensuring good accuracy, improves real-time performance, and solves the problem of having to test each pair of points to find the optimal partition and combination structure when clustering and pairing key features.

[0045] Regarding the lightweighting of the model, MoveNet is a bottom-up key feature detection model for actor skeletons launched by Google, which uses heat maps to accurately locate the key features of actors. APIs and models such as TensorFlow Lite are provided and can be deployed on web pages and mobile phones. The architecture consists of two parts: a feature extractor and a set of prediction heads.

[0046] The feature extractor in MoveNet is MobileNetV2, with an additional Feature Pyramid Network (FPN), which can output feature maps with high resolution (output stride of 4) and rich semantics. Four prediction heads are attached to the feature extractor, which are responsible for dense prediction: the actor center heat map, the key feature regression field, the actor key feature heat map, and the 2D offset field for each key feature.

[0047] Based on the above prediction calculation method, the following operation sequence can be followed:

[0048] 1. The actor center heat map is used to identify the centers of all individuals in the frame, which is defined as the arithmetic mean of all key features belonging to the individual. The position with the highest score is selected (weighted by the inverse distance from the center of the frame).

[0049] 2. The initial key feature set of the actor is generated by segmenting the key feature regression output corresponding to the pixel at the object center. Since this is a center-outward prediction (which must operate at different scales), the quality of the regression key features is not particularly accurate.

[0050] 3. Each pixel in the key feature heat map is multiplied by a weight that is inversely proportional to the distance of the corresponding regression key feature. This ensures that we do not accept key features from background figures because they are usually not close to the regression key features and thus have lower scores.

[0051] 4. The final set of key feature predictions is selected by retrieving the coordinates of the maximum heat map value in each key feature channel. Then the local 2D offset predictions are added to these coordinates to give an accurate estimate.

[0052] After extracting multiple physical process features during the traceability process and achieving efficiency improvement in the traceability process, the calculation alignment for the feature stream is a crucial key algorithm. Specifically: 1) At the data source level, the DTW (Dynamic Time Warping) technical route is used to sort out the anomalies in the time series data that appear in the physical acquisition device; 2) By designing a heterogeneous channel quality function in the Beidou or synchronous communication integrated forwarding device to automatically determine the channel quality, the data after DTW sorting is sent with different priority data packets for data aggregation according to the channel quality; 3) After data aggregation, at the database management level, efficient collaborative management of various feature data is achieved based on a distributed unstructured database.

[0053] DTW time series processing technology. A time series is a data sequence arranged in chronological order, with equal distances maintained between data points. As one of the most common representations of data, time series often need to be matched and compared to calculate the similarity between two different sequences. In actual scenarios, the lengths of two time series often differ or the sequences have drifts on the time axis. For a large number of data nodes, some node data usually needs to be combined and stored. However, due to communication transmission problems or sensor failures, the collected information is usually incomplete, resulting in missing information on the time axis. At this time, DTW is used to align two or more groups of time series, so that the missing segments will not affect the overall data storage.

[0054] The DTW (Dynamic Time Warping) algorithm aligns the time axis based on the method of dynamic programming, eliminating the errors caused by sequence length and misalignment, and obtaining a smaller distance. It allows sequences to be paired by self-replication on the time axis. One data point in sequence A can correspond to multiple data points in sequence B, and conversely, one data point in sequence B can also correspond to multiple data points in sequence A, and the correspondence of data points does not require alignment on the time axis. Through this regularization on the time axis, the points on one sequence are paired with the most similar points on another sequence. Essentially, it is the scaling and offset of the time axis, thus solving the comparison of time series with different lengths or two time series with offsets on the time axis. And the corresponding relationship between the data points on the two time series is the regularization path, which is represented as the "dotted line" connecting the points between Figure 5 and Figure 6 as the "dotted line" connecting the points.

[0055] The calculation process of the DTW distance is as follows:

[0056] Suppose there are two time series q and c with lengths m and n respectively;

[0057] q[1:m] = {q_1, q_2, …, q_m};

[0058] c[1:n] = {c_1, c_2, …, c_n};

[0059] The calculation steps first construct a distance matrix of size m×n. The match between (i, j) (1 ≤ i ≤ m, 1 ≤ j ≤ n) in the matrix corresponds to the data points qi and cj, and the value of this element represents the distance between the two points, which is called the base distance.

[0060] After constructing the distance matrix, the corresponding relationship between the (q, c) point pairs is manifested as a regular path from the square (1, 1) to the square (m, n) in the bending matrix. w i = f w (q i , c j ) represents the pairing of the i-th point in the q sequence and the j-th point in the c sequence, and this point pair is denoted as w i . w i The entire set W of w is the set of all matches of the q and c sequences, that is, the regular path;

[0061] W = W1, W2, …, W1 (max(m, n) ≤ l ≤ m + n).

[0062] The traditional DTW algorithm has a high complexity of O(N2). When both time series are relatively long, the DTW algorithm is relatively slow and cannot meet the requirements. The method in this study uses FastDTW to accelerate the calculation of DTW by restricting the search space and data abstraction, which is mainly divided into three steps:

[0063] Coarse-graining: First, perform data abstraction on the original sequence. Data abstraction can be performed multiple times. For example: 1 / 1 → 1 / 2 → 1 / 4 → 1 / 16. The abstracted coarse-grained data points are the averages of multiple fine-grained data points;

[0064] Projection: Use the DTW algorithm on a coarser granularity to obtain a regular path;

[0065] Fine-graining: Fine-grain the path obtained by running the DTW algorithm on the coarse-grain and the space within a certain range around it. Usually, expand K grains around the path. K is the radius parameter, generally taken as 1 or 2.

[0066] The algorithm first performs DTW operations on a coarser granularity (1 / 8) dimension. Then, fine-grain the obtained regular path, determine the number of fine-grained units expanded outward through the radius parameter, and then execute the DTW algorithm to obtain a new regular path. And so on, continuously fine-graining. Because the search space is restricted, the time complexity of the FastDTW algorithm is relatively low, which is O(N). Through the matching of DTW, the time axes of different sensors are aligned with each other, and the specific effects are as follows Figure 11 as shown.

[0067] Example 3

[0068] The secondary fusion process of unstructured data and structured data is to map unstructured feature data into structured Key keys by introducing unstructured key - value pairs. The key - value pairs adopt a unified hash mapping method. For example, using hash256 key - values, under the requirement of irreversible mapping ability, the unstructured data is mapped to unified structured data. On this basis, traditional structured data such as two - dimensional codes is also uniformly mapped to the above HASH256 values, providing a simplified secondary structured WMS data structure. The specific steps are as follows:

[0069] Step 1: Physical process feature data without direct physical meaning will be mapped into unified key - values through processes such as hash mapping;

[0070] Step 2: The structured data in the existing logic layer will be mapped into unified key - values through processes such as hash mapping;

[0071] Step 3: The system will form secondary structured data with consistent length and structured height for the unified mapped key - values, providing input conditions for subsequent credible traceability.

[0072] Example 4

[0073] After the automated extraction of various physical process feature data and the execution of the secondary fusion steps of unstructured data and structured data, the present invention will construct a blockchain - based physically credible distributed inventory system for the WMS application scenario. As described above, physical credibility means that the unstructured features of the physical process are collected by the neural network structure (possibly without corresponding physical meaning) and jointly complete a unified irreversible hash256 mapping with the structured data in the existing logic layer.

[0074] In this link, each virtual WMS data management object is realized through the unified mapped structured data. When the traceability process is required, the system cannot directly give the traceability regression result, but can only give the classification result, that is:

[0075] In the direct mapping relationship of the feature data, since the irreversible mapping has been completed, in the direct interpretation process of the data, the original traceability information cannot be directly obtained, and only the comparison result of whether the traceability is correct can be obtained;

[0076] To obtain the original traceability data of the real physical process, it is necessary to find the corresponding original data in the original database according to the steps of the secondary fusion process of unstructured data and structured data by the unique key - value pair, and thus conduct the traceability of the real data source for physical credibility traceability;

[0077] When the system uses neural network characterization parameters for physical process feature extraction, whether it can complete the reversible analysis of the above-mentioned physically credible traceability depends on whether the involved WMS system retains the original machine vision images, and the uniqueness of its accuracy depends on whether the distributed vision data is subjected to secondary data fusion under a unified time and space benchmark;

[0078] According to the above steps, an internal fast indexing mechanism can be established to accelerate the reverse mapping of the original data for the above-mentioned physically credible traceability.

[0079] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A physical trusted traceability warehousing management device based on unstructured blockchain features, characterized in that Including: Through machine vision methods, automatically extract various physical process characteristic data for WMS application scenarios; The secondary fusion process of unstructured data and structured data; For WMS application scenarios, build a physically trusted distributed inventory system based on blockchain, and give classification results during the traceability process; The automatic extraction process of physical process characteristic data includes the following steps: Step 1: Through the neural network structure used in the machine learning method of machine vision, automatically extract features from unstructured data, including the mapping of unstructured data on the physical behavior of operators, the information changes in the physical process of goods transfer, and the changes in target shelves; Step 2: The above mapping data is regarded as the internal representation parameters of the neural network. Without analyzing its explicit physical meaning, it is directly regarded as the extraction result of physical process characteristics to extract physical process characteristic data without direct physical meaning; Step 3: The extraction results of the above physical processes can be aligned and fused with distributed multi-nodes under the time and space benchmarks of the WMS system; In Step 3, the fusion process is simple feature superposition, feature semantic fusion, or feature secondary analysis fusion; The unstructured data of physical layer features will be fused with the structured data of the Internet of Things at the upper layer of the cloud platform logic layer; The secondary fusion process of the unstructured data and the structured data is to map the unstructured feature data into a structured Key key by introducing unstructured key-value pairs.

2. The physical trusted traceability warehousing management device based on unstructured blockchain features according to claim 1, wherein The key-value pairs adopt a unified hash mapping method, providing a simplified secondary structured WMS data structure. The specific steps are as follows: Step 1: The physical process characteristic data without direct physical meaning will be mapped into a unified key-value through the hash mapping process; Step 2: The existing structured data in the logic layer will be mapped into a unified key-value through the hash mapping process; Step 3: The system will form secondary structured data with the same length and structured height for the unified mapped key-values, providing input conditions for subsequent trusted traceability.

3. The physical trusted traceability warehousing management device based on unstructured blockchain features according to claim 1, wherein The physically trusted distributed inventory system realizes each virtual WMS data management object through the unified mapped structured data.

4. The physical trusted traceability warehousing management device based on unstructured blockchain features according to claim 1, characterized in that, During the physically trusted traceability process, by establishing an internal fast indexing mechanism, the reverse mapping of the original data for physically trusted traceability can be accelerated.

Images