Tobacco logistics industry global intelligent supervision platform system

Abstract

The invention relates to the technical field of logistics, and discloses a tobacco logistics industry global intelligent supervision platform system, which adopts YoLo, ResNet, multipath, quantum parallel, quantum Fourier transform, opencv, OCR / SIFT skeleton key point detection and deepsort tracking technologies to carry out image recognition and state recognition on monitoring video information and recognize markers in an environment. The identification result obtained after identification is analyzed, the number of the strip-shaped cigarettes in the picture, the ID appearance duration, the positions of the strip-shaped cigarettes, the loading and unloading conditions and the label conditions are determined and calculated, whether violent loading and unloading are carried out or not is judged, the falling angle and the falling condition of the strip-shaped cigarettes are judged, an alarm is given, and a red light is displayed. The processing result and the corresponding image or video are sent back to the intelligent platform display device; and the intelligent platform display device is used for receiving the content transmitted by the data processing module and storing the data for platform management personnel to perform management and information pushing.

Description

Technical Field

[0001] The present invention relates to the field of logistics technology, and in particular to a global intelligent supervision platform system for the tobacco logistics industry. Background Art

[0002] Lean management in the tobacco logistics industry requires efficient transportation of tobacco to the corresponding customers and guaranteed quality of goods. Traditional manual methods usually use manual monitoring of the discharge port, transmission process, labeling status, delivery process, and product receipt status.

[0003] However, this traditional manual method requires employees to pay great attention and energy, and to deploy specialized personnel at specific locations in the corresponding areas. The larger the amount of tobacco, the more personnel are required to be deployed, resulting in high management costs for the company. In addition, there are many mistakes in personnel monitoring during work, and it is impossible to concentrate highly throughout the process, and it is easily disturbed by many environmental factors. In addition, there is no real-time control method for online logistics products, the environment for retail delivery and receipt is difficult to verify, manual review is inefficient, label status is unregulated, closed-loop management is difficult, and full-process supervision is non-existent.

[0004] The resulting loose management of tobacco logistics and imprecise management of tobacco may lead to certain problems and ultimately a decline in overall operations. Summary of the invention

[0005] The purpose of the present invention is to solve the above problems and to design a global intelligent supervision platform system for the tobacco logistics industry.

[0006] The present invention provides a global intelligent supervision platform system for tobacco logistics industry, and the global intelligent supervision platform system for tobacco logistics industry comprises the following steps:

[0007] An information collection device is used to collect monitoring video information in the tobacco factory using a high-resolution camera, and send the collected monitoring video information to a data processing module through a communication interface, wherein the monitoring video information at least includes tobacco video information, staff video information and environment video information;

[0008] The data processing module is used to use YoLo, ResNet, multi-channel, quantum parallel, quantum Fourier transform, opencv, OCR / SIFT skeleton key point detection and deepsort tracking technology to perform image recognition, state recognition and environmental marker recognition on the monitoring video information, analyze the recognition results obtained after recognition, determine and calculate the number of cigarette strips in the picture, the duration of ID appearance, the position of the cigarette strips, the loading and unloading conditions, and the labeling conditions, determine whether violent loading and unloading is performed, and the falling angle and condition of the cigarette strips, and issue an alarm and display a red light, and send the processing results and the corresponding images or videos back to the smart platform display device;

[0009] The smart platform display device is used to receive content transmitted from the data processing module and store data for management and information push by platform managers.

[0010] Optionally, in a first implementation of the present invention, the data processing module specifically includes the following process:

[0011] The number of smoke strips in the monitoring video information screen and the duration of the ID appearance are calculated through deepsort algorithm, multi-channel and quantum parallelism, and quantum Fourier transform;

[0012] Use Yolo target detection, multi-channel and quantum parallelism, and quantum Fourier transform to determine the position of the smoke strip;

[0013] The position, scale and direction of the key points are extracted by OCR / SIFT, the distance is calculated by Euclidean distance, Hamming distance and cosine distance, the local features are matched, and the SIFT features are extracted to analyze the label situation in the monitoring video information screen;

[0014] The activation mechanism identifies the detected effective identification features, calculates the state of the falling smoke strip in the picture through deepsort tracking, opencv, vector, and Huffman transform, and uses time series detection, behavior detection, violent algorithm detection, and skeleton detection to determine whether there is violent behavior, wherein the effective identification features at least include roads, carriages, carts, and bare soil on the ground;

[0015] Through YoLo and ResNet algorithms, landmarks in the environment are identified and the identified images are classified.

[0016] Optionally, in a second implementation of the present invention, the position, scale, and direction of the key points are extracted by OCR / SIFT, the distance is calculated by Euclidean distance, Hamming distance, and cosine distance, local features are matched, SIFT features are extracted, and the label situation in the monitoring video information screen is analyzed, including:

[0017] The OCR / SIFT algorithm is used to describe the position, scale and direction of key points, and the grayscale images between adjacent Gaussian scales of the same size are subtracted to obtain Gaussian difference images;

[0018] Detect feature points on the Gaussian difference image, summarize them after extraction, and calculate the Euclidean distance between feature description operators:

[0019]

[0020] If the distance is less than the threshold, the match is successful:

[0021]

[0022] Compute the distance between vectors:

[0023]

[0024] Optionally, in a third implementation of the present invention, the method of determining the position of the smoke strip by using Yolo target detection, multi-path and quantum parallelism, and quantum Fourier transform includes:

[0025] The quantum superposition principle of the multi-channel algorithm model is used, Yolov5 is used to read multiple cameras, and the morphological edge extraction algorithm of quantum superposition state and information entropy is used to perform real-time analysis and early warning of the monitoring screen;

[0026] The model is adaptively expanded and trained through the active learning strategy of entropy bagging query, the sample entropy is calculated, and samples are selected for labeling based on the entropy value. The entropy value calculation formula is as follows:

[0027]

[0028] Optionally, in a fourth implementation of the present invention, the activation mechanism identifies the detected effective identification features, and calculates the state of the falling of the strip smoke in the picture through deepsort tracking, opencv, vector, and Huffman transform, including:

[0029] The Canny operator of opencv is used to detect the edge of the image, and the Huffman transform is used to find all the point and line references. The Huffman exchange formula is as follows:

[0030] lines=cv.HoughLines(image, rho, theta, threshold[, lines[, srn[, stn[, min_theta[, max_theta]]])

[0031] According to Huffman exchange, a three-dimensional space is established for the video or picture to form a vector, and then the vector is used to calculate and analyze the angle. According to the angle calculation formula, it is determined whether the cigarette package has a corner or edge touching the ground. The angle calculation formula is:

[0032]

[0033] The Huffman transform is used to extract the feature vector a of the cigarette pack, and the angle between it and the direction vector b of the edge in the image is calculated. If cos(θ)≤0.5, it means that the cigarette pack may have touched the ground at a corner or edge, thus inferring the state of the cigarette strip when it fell to the ground and whether it is damaged.

[0034] Optionally, in a fifth implementation of the present invention, the use of timing detection, behavior detection, violent algorithm detection, and skeleton detection to determine whether there is violent behavior includes:

[0035] The video depth image is obtained through the Kinect somatosensory device, and the captured image is converted into a joint node image and the coordinate information of the bone node is obtained;

[0036] The human joint movement patterns of violent sorting behavior are analyzed, and the joint movement angle speed and coordinate changes are taken as target features. The Top-Down or Bottom-Up method is used to detect and judge whether there is violent behavior by fusing the facial features, hand joint video and sound features, and image features.

[0037] Optionally, in a sixth implementation of the present invention, the information collection device includes at least a camera and a portable camera device, the camera is fixedly installed at the loading and unloading personnel, and detects the situation when the tobacco is on the track; the portable camera device is installed on the staff to take photos or videos of the tobacco and the staff in the tobacco factory.

[0038] Optionally, in a seventh implementation of the present invention, the smart platform display device includes a login interface, a user management interface, a parameter panel, and a log management interface.

[0039] The login interface has a telescopic function, and the display mode can be freely selected. Videos can be selected in different areas;

[0040] User management interface, for free adding, deleting, modifying and exporting;

[0041] Parameters section, for real-time improvement and maintenance;

[0042] The log management interface is used to manage and delete administrator user logs.

[0043] Optionally, in the eighth implementation of the present invention, the login interface is divided into system management, system monitoring and intelligent perception sections with different bars; the intelligent perception section groups and modifies devices, detects abnormal alarms and pushes them to platform managers.

[0044] Combining all the above technical solutions, the advantages and positive effects of the present invention are as follows:

[0045] 1. In terms of cost, it has changed the traditional manual inspection mode, saving 2 people for each set of equipment, and the company can reduce the human supervision cost by 70% on average;

[0046] 2. In terms of recognition accuracy, manual inspection feedback and single machine vision inspection optimization are promoted to the data management platform to fully improve system performance. The hardware and software of the machine vision algorithm module are combined to perform integrated inspection with higher accuracy.

[0047] 3. In terms of efficiency, the all-weather, blind-angle-free intelligent detection system realizes real-time detection, immediate response, and precise closed-loop, realizes real-time analysis and early warning of monitoring images and transmits them to the command center for decision-making and processing, and the AI ​​model response speed is less than 500ms; it realizes accurate recognition and trigger detection of key elements in the scene, and the detection mechanism will be activated only when the three key targets of shelves (including cigarette packs), personnel, and trucks appear at the same time, reducing server resource consumption by 20%; the studied violence detection algorithm accurately analyzes the dynamics in the scene through the dual dimensions of personnel skeleton point posture estimation and cigarette pack landing posture, and quickly determines whether it is a potential violent loading behavior, and the model recognition accuracy rate is ≥99.5%; it can learn new behavioral characteristics of personnel loading according to environmental changes and extract personnel behavior characteristics from the video without intervention, and complete the self-update of the database through incremental learning; deep analysis of images and videos, automatic detection and analysis of the delivery site environment to accurately determine whether the delivery location and method are compliant. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the following detailed description of the preferred embodiment.The drawings are only for the purpose of illustrating the preferred embodiments and are not to be construed as limiting the invention.

[0049] Figure 1 A process flow chart of the global intelligent supervision platform system for the tobacco logistics industry provided by an embodiment of the present invention;

[0050] Figure 2 A technical flow chart of the global intelligent supervision platform system for the tobacco logistics industry provided by an embodiment of the present invention;

[0051] Figure 3A schematic diagram of monitoring the activation mechanism process provided by an embodiment of the present invention;

[0052] Figure 4 A schematic diagram of strip smoke morphology detection provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0054] Example 1.

[0055] See also Figure 1 and Figure 2 As shown, this embodiment provides a tobacco logistics industry global intelligent supervision platform system including the following steps:

[0056] An information collection device is used to collect monitoring video information in the tobacco factory using a high-resolution camera, and send the collected monitoring video information to the data processing module through a communication interface, wherein the monitoring video information includes at least tobacco video information, staff video information and environment video information;

[0057] The data processing module is used to use YoLo, ResNet, multi-channel, quantum parallel, quantum Fourier transform, opencv, OCR / SIFT skeleton key point detection and deepsort tracking technology to perform image recognition, state recognition and environmental marker recognition on the monitoring video information, analyze the recognition results obtained after recognition, determine and calculate the number of cigarette strips in the picture, the duration of ID appearance, the position of the cigarette strips, the loading and unloading situation, and the label situation, determine whether violent loading and unloading is performed, and the falling angle and situation of the cigarette strips, and issue an alarm and display a red light, and send the processing results and the corresponding images or videos back to the smart platform display device;

[0058] The smart platform display device is used to receive content transmitted from the data processing module and store data for management and information push by platform managers.

[0059] In this embodiment, the information acquisition device uses high-resolution cameras (both fixed and track) to capture images or videos of tobacco in the tobacco factory and images or videos of the staff, and record relevant environmental information; the camera can be fixedly installed at the loading and unloading personnel, and detect the situation when the tobacco is on the track; maintenance personnel can use portable cameras to take photos or videos of tobacco and staff in the tobacco factory, and upload them to the information platform for further processing and analysis; the information acquisition device sends the collected image or video data to the data processing module through the communication interface.

[0060] In this embodiment, the data processing module adopts YoLo, ResNet, multi-channel, quantum parallel, quantum Fourier transform, opencv, OCR / SIFT skeleton key point detection and deepsort tracking technology to enable the server to obtain the data sent by the information collection device; computer vision technology is used on the server to perform image recognition, state recognition, and identification of markers in the environment or direct image classification and other processing and analysis; by analyzing the processed data, the number of cigarette strips in the picture, the duration of ID appearance and other information, the location of the cigarette strips, loading and unloading conditions, and labeling conditions are determined; it is determined whether violent loading and unloading is carried out, and the falling angle and condition of the cigarette strips are alarmed and red lights are displayed, and the processing results and corresponding images or videos are sent back to the smart platform display device.

[0061] In this embodiment, the smart platform display device receives data from the data processing module and some data stored in the background, and provides a platform to detect tobacco and staff conditions. The interface is an example of an embodiment of the invention, which is a tobacco logistics supervision and identification interface, specifically: the platform is first a login interface, which is divided into system management, system monitoring, and intelligent perception sections with different bars. It has a telescopic function, and the display mode can be freely selected. Videos can choose videos in different areas. The user management interface can be freely added, deleted, modified, exported, etc. The parameter section can be improved and maintained in real time. The user logs of managers can be managed and deleted, etc. In the intelligent perception section, devices can be grouped and modified, and modification algorithms can be added to meet the algorithm and optimization when changes occur later. At the same time, it also has abnormal alarm reminders and push notifications to platform managers. The above is only one example of this example, and the specific example of the invention can be changed according to user needs.

[0062] In this embodiment, the algorithm uses small sample data model expansion technology to solve the problem of small amount of training data caused by uncontrollable human behavior, cigarette status, and labeling conditions. It uses flipping, rotation, cropping and other methods to amplify the training data, and extracts the key attributes of the sample through a semantically constrained adversarial network to achieve data enhancement.

[0063] A tobacco logistics monitoring system is used. The system uses deepsort tracking and Yolo algorithm and analyzes in real time based on video monitoring whether the equipment has blocked material ports, hanging cigarettes, or dropped cigarettes during the process of spitting out cigarettes. Deepsort tracking first uses a target detector to determine the position of the cigarette box in each frame of the video, and then uses a deep learning model to extract the appearance features of the cigarette box. Subsequently, through technologies such as the Hungarian algorithm and Kalman filtering, it realizes continuous tracking of multiple cigarette boxes, and then performs cascade matching - it first tries to match the detection results with high-confidence trajectories, and then matches with low-confidence trajectories, and finally performs trajectory management - maintains the trajectory of each target, and initializes a new trajectory for the newly detected target. In this way, the specific situation of the cigarette box can be judged, and an alarm response can be made in a timely manner. Measures are taken on both the workbench and the computer to remind the staff.

[0064] Using the OCR / SIFT algorithm, first of all, the position of the key point, i.e. the coordinates, the scale of the key point, if the scale information is not extracted, the description of the feature point will not be based on the scale information, and thus it will not have scale invariance. The direction of the key point, if the direction information is not extracted, the description of the feature point will not be based on the direction information, and thus it will not have rotation invariance. After extracting these three pieces of information, local matching is performed, and the SIFT feature point position and scale extraction is performed. The grayscale images between adjacent Gaussian scales of the same size are subtracted to obtain a Gaussian difference image. This makes the edge features more obvious. Then the feature points are detected on the Gaussian difference image, and after the extraction is completed, they are summarized. The distance between the vectors is calculated using formula 1. When the distance is less than a certain threshold, it is considered that the two feature points match, that is, the match is successful. In this way, the label situation can be judged, whether there is missing printing, overprinting, or missing printing; feature points are detected on the Gaussian difference image, and after the extraction is completed, they are summarized, and the Euclidean distance between feature description operators is calculated:

[0065]

[0066] If the distance is less than the threshold, the match is successful:

[0067]

[0068] Compute the distance between vectors:

[0069]

[0070] The quantum superposition principle of the multi-channel algorithm model is used, and Yolov5 is used for multi-channel camera reading. The morphological edge extraction algorithm of quantum superposition state and information entropy can improve the efficiency of image processing, realize real-time analysis and early warning of the monitoring screen and transmit it to the command center for decision-making and processing. The AI ​​model response speed is less than 500ms, and it can also detect the lack of printing, overlapping and missing printing of labels.

[0071] The model is adaptively expanded and trained through the active learning strategy of entropy bagging query. The sample entropy is calculated according to Formula 2. Representative samples are selected for manual labeling based on the entropy value. The model is quickly established to achieve adaptive expansion training, which greatly reduces the time and workload of manual labeling and improves the recognition accuracy. The entropy value calculation formula is:

[0072]

[0073] The logistics behavior feature detection algorithm for identifying loading behavior of personnel is adopted. Through the algorithm, accurate identification and trigger detection of key elements in the scene are achieved. The detection mechanism will be activated only when the three key targets of shelf (including cigarette packs), personnel and trucks appear at the same time, which reduces the server resource consumption by 20%. The activation mechanism process is as follows: Figure 3 As shown:

[0074] A violent detection algorithm is used to detect violent loading and unloading of personnel. First, the behavior state of the person is judged by using time sequence detection and behavior detection. The video depth image is obtained through the Kinect somatosensory device, and the captured image is converted into a joint node image and the bone node coordinate information is obtained. Based on this information, the human joint movement pattern of violent sorting behavior is analyzed, and the joint movement angle speed and coordinate change are used as target features. A fusion model of joint node features such as facial features and hand joints is proposed, and video sound features and image features are integrated. The Top-Down or Bottom-Up method is used for detection to make corresponding judgments.

[0075] Then, the cargo shape is detected, the Canny operator of opencv is used to detect the edge of the image, and the Huffman transform is used to find all the point and line references. The Huffman exchange formula is used to calculate to establish a three-dimensional space for the video or picture, so as to form a vector:

[0076] lines=cv.HoughLines(image, rho, theta, threshold[, lines[, srn[, stn[, min_theta[, max_theta]]]]])

[0077] Then use the vector to calculate and analyze the angle, and use the angle calculation formula to determine whether the cigarette package has landed on the corner or edge:

[0078]

[0079] The feature vector a of the cigarette package is extracted by Huffman transform, and the angle between it and the direction vector b of the edge in the image is calculated. If cos(θ)≤0.5, it means that the cigarette package may have a corner or edge touching the ground, so as to infer the state of the cigarette bar falling to the ground and whether it is damaged. In this way, the damage of the cigarette package can be inferred. Figure 4 shown.

[0080] When the violent algorithm detects that there is violent loading and unloading and the cigarette package is damaged, the alarm mechanism will be triggered to alert the staff and the computer at the same time. The model has low complexity and is easy to calculate. The recognition accuracy of the model is ≥99.5%.

[0081] Real-time monitoring of personnel reduces staff errors, lowers costs, is real-time and efficient, and allows 24-hour observation. Data is automatically summarized and counted, which is concise and clear, reducing waste of human resources and saving the process of manual inspection.

[0082] Example 2.

[0083] The present embodiment provides a full-domain intelligent supervision platform for the tobacco logistics industry based on machine vision, which includes the following steps: using various cameras to collect information such as cigarette conditions, manual loading and unloading, and labeling conditions, including monitoring videos and related environmental information; performing target detection on the collected monitoring videos, and using computer vision technology algorithms to analyze whether the discharge port is blocked, cigarettes are hanging, or falling during the smoke-spitting process of the equipment; analyzing the cigarette label printing screen in the picture to see whether the labels are missing, overlapped, or missed; real-time analysis of the identification of cigarette strips on the conveyor belt to see whether the cigarette strips stand up, overlap, or fall; performing personnel behavior recognition, and real-time analysis of the status of employees carrying cigarette strips and the posture of cigarette strips falling to the ground in the picture to determine whether there are people carrying them violently, and the system automatically alarms in the event of the above abnormal conditions; effectively identifying the features in the picture: roads, carriages, carts, bare soil on the ground, etc., and sending alarms when the object features exceed the set threshold to determine that the delivery is unqualified; the platform realizes its innovation and technological advancement, and also demonstrates its application value and broad application prospects in the logistics industry, especially the tobacco logistics industry.

[0084] A machine vision-based global intelligent supervision platform for the tobacco logistics industry includes:

[0085] Step S1: Select a corresponding information collection scheme according to the type and location of the camera in the tobacco factory;

[0086] Step S2: Calculate the number of smoke strips in the image and the duration of the ID appearance by using deepsort algorithm, multi-path and quantum parallelism, and quantum Fourier transform;

[0087] Step S3: The server uses Yolo target detection, multi-channel and quantum parallelism, and quantum Fourier transform to determine the position of the smoke strip;

[0088] Step S4: Analyze the label situation in the picture through OCR / SIFT matching;

[0089] Step S5: The activation mechanism starts to identify effective features. If the features are met, detection begins. The state of the falling smoke strip in the picture is calculated through deepsort tracking, opencv, vector, and Huffman transform. Time series detection, behavior detection, violent algorithm detection, and skeleton detection are used to determine whether there is violent behavior.

[0090] Step S6: Use YoLo and ResNet algorithms to identify landmarks in the environment or directly classify images.

[0091] The step S1 is specifically as follows:

[0092] Step S11: Use various cameras to collect information such as the status of cigarette strips, manual loading and unloading, and labeling.

[0093] Step S12: Select a suitable information collection solution based on the camera type and location to ensure accurate and effective identification and association of different situations.

[0094] The step S2 is specifically as follows:

[0095] Step S21: The collected cigarette status information is transmitted to the server, and the deepsort algorithm is used for data processing and storage, so as to calculate the number of cigarette strips and the duration of ID appearance.

[0096] Step S22: Enhance computer capabilities and save resources through multi-channel, quantum parallelism, and quantum Fourier transform.

[0097] The step S3 is specifically as follows:

[0098] Step S31: Perform Yolo target detection on the collected image or video to determine the position of the cigarette strip;

[0099] Step S32: Enhance computer capabilities and save resources through multi-channel, quantum parallelism, and quantum Fourier transform.

[0100] The step S4 is specifically as follows:

[0101] Step S41: extract the position, scale and direction of the key points through OCR / SIFT, calculate the distance through Euclidean distance, Hamming distance, cosine distance, etc., match the local features, extract the features through SIFT, and analyze the label situation in the matching picture.

[0102] The step S5 is specifically as follows:

[0103] Step S51: Identify the behavior of the personnel and effectively identify the features in the image: road, carriage, cart, bare soil on the ground. Only when the features are met at the same time, the activation mechanism is started and detection begins.

[0104] Step S52: Use timing detection, behavior detection, brute force algorithm detection and human skeleton key point detection algorithm to detect some key points of the human body, such as joints, facial features, etc., describe the human skeleton information through key points, use Top-Down or Bottom-Up method to detect, start detection from the behavior, and timing detection after the data has deviation, through brute force detection and efficient processing, so as to determine whether the staff has unmanned throwing action.

[0105] Step S53: Preprocess the video or picture, extract valid data, perform edge detection on the image through the Canny operator of opencv, calculate all line references through statistical Huffman transform, establish a three-dimensional space for the video or picture, thereby forming a vector, analyze the angle, calculate the angle through the vector, and thus determine the landing angle of the strip smoke.

[0106] Step S54: When the following characteristics are met: whether there is any throwing action by someone, and whether there is any corner situation, only when all the characteristics are met at the same time, it is determined whether there is any person carrying the object violently, and measures such as alarming and displaying a red light are taken. The processing results and the corresponding images or videos are sent back to the smart platform display device.

[0107] The step S6 is specifically as follows:

[0108] Step S61: The delivery person uploads photos in real time, detects the target through YoLo, identifies the landmarks in the environment or directly classifies the pictures, and uses the ResNet algorithm to identify the landmarks in the environment or directly classify the pictures, so as to detect whether the cigarettes are signed for correctly.

[0109] Through the implementation of the above solution, the traditional manual inspection mode has been changed in terms of cost, and each set of equipment saves 2 people, and the company can reduce the labor supervision cost by 70% on average. In terms of recognition accuracy, the manual inspection feedback and the single machine vision inspection optimization are promoted to the data management platform to fully improve the system performance; the hardware and software of the machine vision algorithm module are combined for integrated inspection, which has higher accuracy; in terms of efficiency, the all-weather and blind-angle intelligent inspection system has real-time detection, immediate response, and precise closed loop, as follows:

[0110] 1. The AI ​​intelligent supervision platform system for the tobacco logistics industry has been developed. It innovatively adopts multi-channel algorithm model analysis and parallel computing based on the principle of quantum superposition to achieve real-time analysis and early warning of monitoring images and transmit them to the command center for decision-making and processing. The response speed of the AI ​​model is less than 500ms.

[0111] 2. The advanced algorithm developed in the research realizes accurate recognition and trigger detection of key elements in the scene. The detection mechanism will be activated only when the three key targets of shelves (including cigarette packs), personnel, and trucks appear at the same time, reducing server resource consumption by 20%;

[0112] 3. The violence detection algorithm studied uses the dual dimensions of human skeleton point posture estimation and cigarette bag landing posture to accurately analyze the dynamics in the scene, quickly determine whether it is a potential violent loading behavior, and the model recognition accuracy is ≥99.5%;

[0113] 4. Build an adaptive and automatically updated behavior recognition model that can learn new behavior characteristics of personnel loading vehicles according to environmental changes and extract personnel behavior characteristics from videos without intervention, and complete the self-update of the database through incremental learning;

[0114] 5. The innovative multimodal large model technology studied can deeply analyze images and videos, automatically detect and analyze the delivery site environment to accurately determine whether the delivery location and method are compliant.

[0115] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. The global intelligent supervision platform system for the tobacco logistics industry is characterized by: The tobacco logistics industry global intelligent supervision platform system includes the following steps: An information collection device is used to collect monitoring video information in the tobacco factory using a high-resolution camera, and send the collected monitoring video information to a data processing module through a communication interface, wherein the monitoring video information at least includes tobacco video information, staff video information and environment video information; The data processing module is used to use YoLo, ResNet, multi-channel, quantum parallel, quantum Fourier transform, opencv, OCR / SIFT skeleton key point detection and deepsort tracking technology to perform image recognition, state recognition and environmental marker recognition on the monitoring video information, analyze the recognition results obtained after recognition, determine and calculate the number of cigarette strips in the picture, the duration of ID appearance, the position of the cigarette strips, the loading and unloading conditions, and the labeling conditions, determine whether violent loading and unloading is performed, and the falling angle and condition of the cigarette strips, and issue an alarm and display a red light, and send the processing results and the corresponding images or videos back to the smart platform display device; The smart platform display device is used to receive content transmitted from the data processing module and store data for management and information push by platform managers.

2. The tobacco logistics industry global intelligent supervision platform system according to claim 1, characterized in that: The data processing module specifically The process includes: The number of smoke strips in the monitoring video information screen and the duration of the ID appearance are calculated through deepsort algorithm, multi-channel and quantum parallelism, and quantum Fourier transform; Use Yolo target detection, multi-channel and quantum parallelism, and quantum Fourier transform to determine the position of the smoke strip; The position, scale and direction of the key points are extracted by OCR / SIFT, the distance is calculated by Euclidean distance, Hamming distance and cosine distance, the local features are matched, and the SIFT features are extracted to analyze the label situation in the monitoring video information screen; The activation mechanism identifies the detected effective identification features, calculates the state of the falling smoke strip in the picture through deepsort tracking, opencv, vector, and Huffman transform, and uses time series detection, behavior detection, violent algorithm detection, and skeleton detection to determine whether there is violent behavior, wherein the effective identification features at least include roads, carriages, carts, and bare soil on the ground; Through YoLo and ResNet algorithms, landmarks in the environment are identified and the identified images are classified.

3. The tobacco logistics industry global intelligent supervision platform system according to claim 2, characterized in that: The feature extraction of the position, scale and direction of the key points by OCR / SIFT, distance calculation by Euclidean distance, Hamming distance and cosine distance, local feature matching, SIFT feature extraction and analysis of the label situation in the monitoring video information screen include: The OCR / SIFT algorithm is used to describe the position, scale and direction of key points, and the grayscale images between adjacent Gaussian scales of the same size are subtracted to obtain Gaussian difference images; Detect feature points on the Gaussian difference image, summarize them after extraction, and calculate the Euclidean distance between feature description operators: If the distance is less than the threshold, the match is successful: Compute the distance between vectors:

4. The tobacco logistics industry global intelligent supervision platform system according to claim 2, characterized in that: The method of using Yolo target detection, multi-path and quantum parallelism, and quantum Fourier transform to determine the position of the smoke strip includes: The quantum superposition principle of the multi-channel algorithm model is used, Yolov5 is used to read multiple cameras, and the morphological edge extraction algorithm of quantum superposition state and information entropy is used to perform real-time analysis and early warning of the monitoring screen; The model is adaptively expanded and trained through the active learning strategy of entropy bagging query, the sample entropy is calculated, and samples are selected for labeling based on the entropy value. The entropy value calculation formula is as follows:

5. The tobacco logistics industry global intelligent supervision platform system according to claim 2, characterized in that: The activation mechanism identifies the detected effective identification features, and calculates the state of the falling smoke strip in the picture through deepsort tracking, opencv, vector, and Huffman transform, including: The Canny operator of opencv is used to detect the edge of the image, and the Huffman transform is used to find all the point and line references. The Huffman exchange formula is as follows: lines=cv.HoughLines(image, rho, theta, threshold[, lines[, srn[, min_theta[, max_theta]]]]]) According to Huffman exchange, a three-dimensional space is established for the video or picture to form a vector, and then the vector is used to calculate and analyze the angle. According to the angle calculation formula, it is determined whether the cigarette package has a corner or edge touching the ground. The angle calculation formula is: The Huffman transform is used to extract the feature vector a of the cigarette pack, and the angle between it and the direction vector b of the edge in the image is calculated. If cos(θ)≤0.5, it means that the cigarette pack may have touched the ground at a corner or edge, thus inferring the state of the cigarette strip when it fell to the ground and whether it is damaged.

6. The tobacco logistics industry global intelligent supervision platform system according to claim 2, characterized in that: The method of using time series detection, behavior detection, violent algorithm detection, and skeleton detection to determine whether there is violent behavior includes: The video depth image is obtained through the Kinect somatosensory device, and the captured image is converted into a joint node image and the coordinate information of the bone node is obtained; The human joint movement patterns of violent sorting behavior are analyzed, and the joint movement angle speed and coordinate changes are taken as target features. The Top-Down or Bottom-Up method is used to detect and judge whether there is violent behavior by fusing the facial features, hand joint video and sound features, and image features.

7. The tobacco logistics industry global intelligent supervision platform system according to claim 1, characterized in that: The information collection device includes at least a camera and a portable camera. The camera is fixedly installed at the loading and unloading personnel and detects the situation of tobacco when it is on the track; the portable camera is installed on the staff to take photos or videos of tobacco and staff in the tobacco factory.

8. The tobacco logistics industry global intelligent supervision platform system according to claim 1, characterized in that: The smart platform display device includes a login interface, a user management interface, a parameter panel, and a log management interface. The login interface has a telescopic function, and the display mode can be freely selected. Videos can be selected in different areas; User management interface, for free adding, deleting, modifying and exporting; Parameters section, for real-time improvement and maintenance; The log management interface is used to manage and delete administrator user logs.

9. The tobacco logistics industry global intelligent supervision platform system according to claim 8, characterized in that: The login interface is divided into system management, system monitoring and intelligent perception sections with different bars; the intelligent perception section groups and modifies devices, detects abnormal alarms and pushes them to platform managers.

Images