A zynq-based intelligent management and monitoring system for production workshop

Abstract

This invention relates to a ZYNQ-based intelligent management and monitoring system for production workshops, belonging to the fields of artificial intelligence and embedded systems. The system includes: an image acquisition module, a ZYNQ system board, and a terminal device connected in sequence; the image acquisition module is used to acquire images of the production workshop in real time; the ZYNQ system board is used to acquire ZYNQ executable target detection files, use these files to detect the production workshop images, and output the detection results; the detection results include whether workers in the production workshop are wearing dust caps, masks, and work uniforms, and whether there are flames in the production workshop; the terminal device is used to display the detection results and issue alarms based on them. This invention replaces a PC with a ZYNQ system board, reducing production costs and system size, and allowing the ZYNQ-based intelligent management and monitoring system for production workshops to be deployed anytime, anywhere, expanding its application scope.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and embedded systems, and in particular to a ZYNQ-based intelligent management and monitoring system for production workshops. Background Technology

[0002] As China vigorously develops its productive forces, accidents in production workshops are also increasing, leading to greater emphasis on hygiene and safety. Many of these accidents are preventable. For example, wearing dust caps can prevent hair from falling onto parts or machinery, thus avoiding product defects and equipment damage; uniform checks can prevent unauthorized personnel from entering the workshop, preventing accidents and losses; and flame detection can detect flames immediately and take timely action to reduce the risk of fire spread. Therefore, intelligent management and monitoring in production workshops have become increasingly important, even indispensable tools.

[0003] However, existing intelligent management and monitoring systems for production workshops are implemented on PCs (personal computers) and placed in fixed monitoring rooms. To meet the real-time monitoring needs of production workshops, a large number of high-performance PCs need to be deployed on the server side to run highly complex target detection algorithms and process image data from multiple production workshops in real time. Although PCs offer excellent detection accuracy and speed, the increased power consumption, high server requirements, and high cost make them unsuitable for large-scale deployment in production workshops. Moreover, implementing a production workshop safety monitoring system on a PC has disadvantages such as poor flexibility and large size, making it unsuitable for deployment to temporary work sites. Summary of the Invention

[0004] The purpose of this invention is to provide a production workshop intelligent management and monitoring system based on ZYNQ, so as to solve the problems of high cost and limited application scope of existing production workshop intelligent management and monitoring systems.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A production workshop intelligent management and monitoring system based on ZYNQ includes: an image acquisition module, a ZYNQ system board, and terminal equipment connected in sequence;

[0007] The image acquisition module is used to acquire images of the production workshop in real time;

[0008] The ZYNQ system board is used to acquire ZYNQ executable target detection files, use the ZYNQ executable target detection files to detect the production workshop images, and output the detection results; the detection results include whether the workers in the production workshop are wearing dust caps, masks and work uniforms, and whether there are flames in the production workshop;

[0009] The terminal device is used to display the detection results and issue an alarm based on the detection results.

[0010] Optionally, the ZYNQ system board includes an SD card for storing the ZYNQ acceleration system and the ZYNQ executable target detection file;

[0011] The ZYNQ acceleration system, located on the SD card, calls the ZYNQ executable target detection file; the ZYNQ acceleration system is also used to accelerate the detection process.

[0012] Optionally, the ZYNQ system board further includes: a processing system terminal, a programmable logic terminal, a USB interface, an Ethernet interface, a display interface, and a serial port; the processing system terminal is connected to the programmable logic terminal; the processing system terminal is also connected to the USB interface, the Ethernet interface, the display interface, and the serial port respectively via an AXI communication bus;

[0013] The processing system terminal is the main control terminal of the ZYNQ system board, used to provide clocks for the programmable logic terminal, the USB interface, the Ethernet interface, the display interface and the serial port;

[0014] The programmable logic terminal is used to control the USB interface, the Ethernet interface, the display interface, and the serial port;

[0015] The USB interface is used to connect the image acquisition module;

[0016] The Ethernet interface is used to connect to the host computer;

[0017] The display screen interface is used to connect to the display screen of the terminal device;

[0018] The serial port is used for serial communication.

[0019] Optionally, the terminal device includes a base, a display screen, and an alarm;

[0020] The display screen and the alarm are connected; the display screen and the alarm are mounted on the base;

[0021] The display screen is used to display the detection results, and the alarm is used to issue an alarm based on the detection results.

[0022] Optionally, the ZYNQ system board is disposed inside the base.

[0023] Optionally, the ZYNQ system board also includes a power on / off button.

[0024] Optionally, the ZYNQ executable object detection file includes the structural information and weight parameters of a trained YOLOv4-tiny network; the ZYNQ executable object detection file is obtained through the host computer, specifically including:

[0025] Prepare a training dataset; the training dataset includes original images of the production workshop and labeled images of the production workshop.

[0026] The YOLOv4-tiny network is trained using the training dataset to obtain the trained YOLOv4-tiny network;

[0027] Save the weight parameters and structure information of the trained YOLOv4-tiny network as a floating-point model file;

[0028] The floating-point model file is quantized to obtain a fixed-point model file;

[0029] The fixed-point model file is converted into the ZYNQ executable target detection file.

[0030] Optionally, the ZYNQ executable target detection file also includes an instruction set for the DPU-accelerated YOLOv4-tiny network and the main program for ZYNQ operation.

[0031] Optionally, the YOLOv4-tiny network includes a backbone feature extraction network, a feature pyramid, and a YOLOhead output layer connected in sequence.

[0032] The backbone feature extraction network is used to extract features from the production workshop image;

[0033] The feature pyramid is used to enhance the feature, resulting in an enhanced feature.

[0034] The YOLOhead output layer is used to detect the enhanced features and obtain the detection results.

[0035] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0036] This invention discloses a ZYNQ-based intelligent management and monitoring system for production workshops. The system acquires images of the production workshop using an image acquisition module. The ZYNQ system board uses ZYNQ executable target detection files to detect the images, checking whether workers are wearing dust caps, masks, and uniforms, and detecting the presence of flames. The system outputs the detection results. A terminal device displays the detection results and issues alarms based on them. This invention replaces a PC with a ZYNQ system board, reducing production costs and system size. The ZYNQ-based intelligent management and monitoring system can be deployed anywhere, expanding its application scope. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 A structural diagram of a ZYNQ-based intelligent management and monitoring system for production workshops provided by this invention;

[0039] Figure 2 This is a schematic diagram of the ZYNQ system board frame provided by the present invention;

[0040] Figure 3 This invention provides a flowchart for creating a ZYNQ executable target detection file.

[0041] Figure 4 The diagram shows the YOLOv4-tiny network structure provided by this invention.

[0042] Figure 5 This is a structural diagram of the Vitis basic acceleration platform provided by the present invention;

[0043] Figure 6 The working principle diagram of the DPU provided for this invention;

[0044] Figure 7 This is a schematic diagram of the ZYNQ acceleration system framework provided by the present invention;

[0045] Figure 8 This is a schematic diagram of the structure of the terminal device provided by the present invention.

[0046] Symbol Explanation: 11. Image Acquisition Module; 12. ZYNQ Executable Target Detection File; 13. ZYNQ Acceleration System; 14. Terminal Device; 15. ZYNQ System Board; 22. Clock Divider; 23. First Resetter; 24. AXI Controller; 25. Interrupt Controller; 26. Second Resetter; 27. Third Resetter; 32. AXI Bus; 33. Command Information; 34. Computation Engine; 35. On-Chip Memory; 36. Memory Controller; 37. On-Chip Memory Buffer; 45. DPU; 46. Constant; 47. Connector; 61. Processing System End; 62. Programmable Logic End; 63. USB Interface; 64. Ethernet Interface; 65. Display Interface; 66. Serial Port; 81. Display Screen; 82. Base; 83. Alarm; 84. Serial Port Opening; 85. Ethernet Opening; 86. USB Opening; 87. Power Button; 89. Power Interface. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] The purpose of this invention is to provide a production workshop intelligent management and monitoring system based on ZYNQ, so as to solve the problems of high cost and limited application scope of existing production workshop intelligent management and monitoring systems.

[0049] The intelligent management and monitoring system for production workshops based on ZYNQ of the present invention aims to solve the problems of high cost, poor portability, high power consumption, and large size of on-site intelligent safety monitoring systems for production workshops implemented on PCs. These problems prevent the on-site intelligent safety monitoring systems for production workshops from being deployed at any time and used on a large scale, thus resulting in a narrow application scope.

[0050] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0051] Figure 1 A structural diagram of a ZYNQ-based intelligent management and monitoring system for production workshops provided by this invention is shown below. Figure 1As shown, the intelligent management and monitoring system for the production workshop based on ZYNQ includes: an image acquisition module 11, a ZYNQ system board 15, and a terminal device 14 connected in sequence. The ZYNQ system board 15 includes an SD card for storing the ZYNQ acceleration system 13 and the ZYNQ executable target detection file 12. The SD card is not stored in... Figure 1 As shown in the image, ZYNQ stands for Zynq-7000 All Programmable SoC, a new generation of fully programmable system-on-chip.

[0052] In practical applications, the ZYNQ acceleration system 13 and the ZYNQ executable target detection file 12 are written together to the SD card of the ZYNQ system board 15, the image acquisition module 11 is connected to the USB interface 63 of the ZYNQ system board 15, and the ZYNQ system board 15 is connected to the terminal device 14.

[0053] The image acquisition module 11 is used to acquire images of the production workshop in real time. In practical applications, the image acquisition module 11 is a USB camera used to acquire image data in real time; this system supports any type of camera.

[0054] The ZYNQ system board 15 is used to acquire the ZYNQ executable target detection file 12. The ZYNQ acceleration system 13, located on the SD card, calls the ZYNQ executable target detection file 12. It uses the ZYNQ executable target detection file 12 to detect images in the production workshop and outputs the detection results. The detection results indicate whether workers in the production workshop are wearing dust caps, masks, and work uniforms, and whether there are flames in the production workshop. The ZYNQ acceleration system 13 is also used to accelerate the detection process. The working principle of the ZYNQ acceleration system 13 calling the ZYNQ executable target detection file 12 is as follows: The ZYNQ acceleration system 13 calls the ZYNQ main program in the ZYNQ executable target detection file 12 to start the entire design process. Then, the ZYNQ acceleration system 13 acquires weight parameters, network structure data, and DPU parallel acceleration instruction sets from the ZYNQ executable target detection file 12 and transmits them to the DPU to achieve parallel acceleration. The YOLOv4-tiny network outputs some data, which is processed by the post-processing code in the main program to obtain the target detection results.

[0055] The terminal device 14 is used to display the detection results and issue an alarm based on the detection results.

[0056] In the production workshop, images are acquired in real time by the image acquisition module 11 and then transmitted to the ZYNQ system board 15 for processing. The ZYNQ system board 15 reads the network structure and weight parameter information of the ZYNQ executable target detection file 12 and transmits it to the ZYNQ acceleration system 13 for parallel acceleration. This system detects in real time whether workers are wearing dust caps, masks, and work uniforms, and performs real-time flame detection. The ZYNQ system board 15 controls other intelligent devices, such as transmitting the detection results to the display screen 81. If any worker is not wearing any of these protective clothing items or if a flame is detected, the target is displayed on the display screen 81 and an alarm is triggered by the alarm 83. This improves the manageability of the production workshop.

[0057] As an optional implementation, the ZYNQ system board 15 further includes: a processing system terminal 61 (PS terminal), a programmable logic terminal 62 (PL terminal), a USB interface 63, an Ethernet interface 64, a display interface 65, and a serial port 66. The structural diagram of the ZYNQ system board 15 is shown below. Figure 2 As shown, the ZYNQ system board 15 is used to control the entire system. The processing system terminal 61 is connected to the programmable logic terminal 62; the processing system terminal 61 is also connected to the USB interface 63, the Ethernet interface 64, the display interface 65, and the serial port 66 via the AXI communication bus. In practical applications, the ZYNQ system board 15 also includes a power button 87, which is used to control the power supply of the ZYNQ system board.

[0058] The processing system terminal 61 is the main control terminal of the ZYNQ system board 15, which provides clock and high-speed AXI communication bus for peripheral interfaces such as the programmable logic terminal 62, the USB interface 63, the Ethernet interface 64, the display interface 65 and the serial port 66.

[0059] The programmable logic terminal 62 is used to control the USB interface 63, the Ethernet interface 64, the display interface 65, and the serial port 66. In practical applications, the PL terminal is used to implement parallel acceleration of algorithms and control peripherals.

[0060] The USB interface 63 is used to connect to the image acquisition module 11 and receive image data (production workshop images) acquired by the image acquisition module 11. The USB interface 63 can also be used as a mouse and keyboard interface to control the ZYNQ system board 15.

[0061] The Ethernet interface 64 is used to connect to the host computer, enabling communication between the ZYNQ system board 15 and the personal computer (host computer) to transmit data or update the system.

[0062] The display interface 65 is used to connect to the display screen 81 of the terminal device 14 and transmit data to the display screen 81 for display.

[0063] The serial port 66 is used for serial communication.

[0064] As an optional implementation, the ZYNQ executable object detection file 12 includes the structural information and weight parameters of the trained YOLOv4-tiny network. The ZYNQ executable object detection file 12 also includes an instruction set for DPU parallel acceleration of the YOLOv4-tiny network and the main program for ZYNQ execution. The main program for ZYNQ execution includes post-processing code for the YOLOv4-tiny network and code that enables ZYNQ to run.

[0065] The ZYNQ executable object detection file 12 is obtained through the host computer. In practical applications, the ZYNQ executable object detection file 12 is obtained through the following steps: training dataset preparation, model training, saving the floating-point model file, quantizing the floating-point model with a quantizer, saving the fixed-point model file, and converting the fixed-point model with a compiler. Figure 3 As shown. The specific steps are as follows:

[0066] Prepare a training dataset; the training dataset includes original images of the production workshop and labeled images of the production workshop. The training dataset is used as input to the model to train the model's weight parameters. Labelimg software is used to label the training dataset.

[0067] The YOLOv4-tiny network is trained using the aforementioned training dataset to obtain a trained YOLOv4-tiny network. In practical applications, model training uses the YOLOv4-tiny network as the training network, trained within the Keras model framework, to train a weight file capable of object detection.

[0068] The weight parameters and structural information of the trained YOLOv4-tiny network are saved as a floating-point model file. In practical applications, the floating-point model file contains the weight parameters of the trained network and information about each layer of the model.

[0069] The floating-point model file is quantized to obtain a fixed-point model file. In practical applications, a quantizer is used to quantize the floating-point model file to reduce memory consumption. This method of model compression is achieved by reducing the computational bit width of the parameters, quantizing a 32-bit floating-point model file into an 8-bit fixed-point model file. The quantizer is an improved quantization model method based on AdaQuant, which does not require retraining the model. It only needs to use a portion of the unlabeled training set to achieve global model optimization and infer the errors between layers, thereby calibrating the activation function and fine-tuning the weight parameters.

[0070] The fixed-point model file is converted into the ZYNQ executable object detection file 12. In practical applications, the compiler is used to convert the fixed-point model file into the ZYNQ executable object detection file 12, and to package the main program running ZYNQ into the ZYNQ executable object detection file 12. After parsing the topology of the output model after the quantizer, an internal computation graph is constructed as an intermediate representation. Based on this, the corresponding control flow and data flow representations are constructed. Subsequently, it performs multiple optimizations, such as computation node fusion (e.g., merging batch normalization into the main convolution), ensuring effective instruction scheduling by utilizing inherent parallelism, or utilizing data reuse, etc.

[0071] As an optional implementation, the YOLOv4-tiny network includes a backbone feature extraction network, a feature pyramid, and a YOLO head output layer connected in sequence. The network structure of the YOLOv4-tiny network is as follows: Figure 4 As shown, the YOLOv4-tiny network is a network used for target detection.

[0072] The backbone feature extraction network is used to extract features from the production workshop images. As a preliminary feature extraction network, each layer consists of Resblock convolutional layers with similar structures.

[0073] The Resblock convolutional layer features a multi-residual structure, effectively mitigating overfitting during model training and enabling rich gradient combinations, making the network easier to optimize. The Resblock convolutional layer first uses a CBL convolutional layer to extract input image features, resulting in large residual edges. Then, the feature map channels are segmented to obtain the backbone and small residual edges. The backbone is then subjected to CBL convolution and stacked with the small residual edges. The stacked feature map is then up-dimensioned and stacked with the large residual edges. Finally, a max-pooling layer compresses the width and height of the feature map.

[0074] The CBL convolution consists of a convolutional layer, a normalized layer, and a Leaky ReLU activation function, forming the entire 2D convolutional structure of the YOLOv4-tiny network.

[0075] The feature pyramid is used to enhance the features, resulting in enhanced features. The feature pyramid is connected to the outputs of the last two layers of the backbone feature extraction network, the Resblock convolutional layer and the CBL convolutional layer. By fusing the features from higher and lower levels, the feature mapping resolution is increased, enabling better detection of targets of different sizes and solving the multi-scale target problem.

[0076] The YOLOhead output layer is used to detect the enhanced features and obtain detection results. Connected to the feature pyramid, the YOLOhead output layer acts as the classifier and regressor of the YOLOv4-tiny network, judging whether an object corresponds to the input feature points. The YOLOhead output layer has three channels that output detection results at different resolutions. Each channel is essentially a 3x3 convolution followed by a 1x1 convolution. The 3x3 convolution integrates features, while the 1x1 convolution adjusts the number of channels.

[0077] In practical applications, the ZYNQ acceleration system 13 includes a Vitis basic acceleration platform, a PateLinux operating system, and a DPU. The Vitis basic acceleration platform provides basic hardware information to the PateLinux operating system and the DPU; the PateLinux operating system provides the ZYNQ system board 15 with an operating system and working environment, enabling it to run the ZYNQ executable target detection file 12; and the DPU accelerates the YOLOv4-tiny network, improving detection speed.

[0078] The Vitis basic acceleration platform is used to prepare for the setup of the PateLinux operating system and DPU. The Vitis basic acceleration platform is built using Vivado 2020.1 software. The Vitis basic acceleration platform is as follows: Figure 5 As shown.

[0079] The processing system terminal 61 serves as the main control terminal of the ZYNQ system board 15, controlling each IP core and peripheral device of ZYNQ, providing a 100MHz clock frequency to each IP core and peripheral device of the ZYNQ system board 15, and providing a high-speed AXI communication bus for data transmission.

[0080] The clock divider 22 is used for clock frequency division and is connected to the clock output port of the processing system terminal 61. It divides the 100MHz clock output by the processing system terminal 61 into 100MHz, 200MHz and 400MHz clocks. The 200MHz and 400MHz clocks are the working clocks prepared for the DPU, and the 100MHz clock is the working clock provided for other IP cores and peripheral devices.

[0081] The first resetter 23 is used to provide reset signals to the IP core and other peripheral devices, and is connected to the three clocks output by the clock divider 22 to obtain reset signals for different operating clocks.

[0082] The AXI controller 24 is used to control the AXI high-speed communication bus and establish communication between various devices, and is connected to the clock output port of the processing system terminal 61 to obtain the working clock.

[0083] The interrupt controller 25 is used to control the interrupt signal generated by the DPU and connect it to the clock output port of the processing system terminal 61 to obtain the working clock.

[0084] Furthermore, the Petalinux system is built on the Ubuntu operating system, and the file containing hardware information generated by the Vitis basic acceleration platform is used as the basic file for building the Petalinux system.

[0085] The working principle of the DPU is as follows: Figure 6 As shown, the DPU45 accelerates the convolutional neural network through instructions and provides accessible memory locations for image data. First, the processing system 61 sends command information 33 to the DPU45 via the AXI bus 32. Upon receiving the command information 33, the DPU45 sends a request to the on-chip memory 35. Then, it retrieves data information from the on-chip memory 35 through the memory controller 36 and caches it in the on-chip memory buffer 37. Finally, the DPU45 uses instructions obtained from the on-chip memory buffer 37 to control the computing engine 34 to perform parallel operations. The computing engine 34 consists of multiple parallelizable processing units.

[0086] Furthermore, such as Figure 7 As shown, the ZYNQ acceleration system 13 is built on the basis of the Vitis basic acceleration platform using Vitis software. The DPU 45 and other devices are built on the Vitis basic acceleration platform. The other devices include the processing system terminal 61, the first resetter 23, the second resetter 26, the third resetter 27, the clock divider 22, the AXI controller 24, the constant 46, the connector 47, and the interrupt controller 25.

[0087] DPU45 is used to accelerate convolutional neural networks. It is connected to clock divider 22 to obtain working clocks at frequencies of 200MHz and 400MHz. It is connected to first reset unit 23, second reset unit 26 and third reset unit 27 to obtain reset signals. It is connected to AXI controller 24 to realize high-speed access to memory data. It is connected to interrupt controller 25 to output interrupt signals to the processing system terminal 61.

[0088] The interrupt outputs of constant 46 and DPU45 are used as inputs to connector 47 to generate a composite interrupt signal.

[0089] The output of connector 47 is connected to interrupt controller 25, and the interrupt signal is fed back to the processing system terminal 61.

[0090] As an optional implementation, the terminal device 14 includes a base 82, a display screen 81, and an alarm 83; the display screen 81 and the alarm 83 are connected; the display screen 81 and the alarm 83 are disposed on the base 82, as shown below. Figure 8 As shown. The display screen 81 is used to display the detection results, and the alarm 83 is used to issue an alarm based on the detection results. The display screen 81 is used to display the real-time status of the production workshop and the marked target information, and is connected to the ZYNQ system board 15 via the display screen interface 65. The alarm 83 is used to issue an alarm sound to indicate the presence of a target not wearing a dust cover, not wearing a uniform, or not wearing a mask, and can also indicate the presence of a flame in the production workshop.

[0091] In practical applications, the ZYNQ system board 15 is housed inside the base 82. Additionally, the terminal device 14 has multiple openings, including a serial port opening 84, an Ethernet opening 85, and a USB opening 86, corresponding to the USB interface 63, Ethernet interface 64, and serial port 66 of the ZYNQ system board 15. A power interface 89 is also provided to supply power to the ZYNQ system board 15.

[0092] This invention discloses a ZYNQ-based intelligent management and monitoring system for production workshops. It employs parallel acceleration technology, accelerating convolutional neural networks by processing multiple pixels of feature maps in parallel. Image preprocessing technology is used to improve the quality of transmitted images, thereby increasing target detection accuracy. Target detection technology utilizes the YOLOv4-tiny target detection algorithm to detect dust caps, masks, work uniforms, and flames in real time and perform further analysis. The YOLOv4-tiny target detection algorithm includes the YOLOv4-tiny network, preprocessing, and post-processing code. Lightweight network model technology is employed to compress the model size of the YOLOv4 target detection network, improving target detection speed and reducing memory usage by reducing the number of network model parameters and computational load. Network model quantization technology is used to reduce the bandwidth of network model weight parameters, thereby reducing the utilization of ZYNQ hardware resources. An energy-efficient hardware design architecture is adopted to achieve energy saving and low power consumption.

[0093] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0094] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A production workshop intelligent management and monitoring system based on ZYNQ, characterized in that, include: The image acquisition module, ZYNQ system board, and terminal device are connected in sequence. The image acquisition module is used to acquire images of the production workshop in real time; The ZYNQ system board is used to acquire ZYNQ executable target detection files, use the ZYNQ executable target detection files to detect the production workshop images, and output the detection results; the detection results include whether the workers in the production workshop are wearing dust caps, masks and work uniforms, and whether there are flames in the production workshop; The ZYNQ executable object detection file includes the structural information of the trained YOLOv4-tiny network and the weight parameters of the trained YOLOv4-tiny network; The ZYNQ executable target detection file is obtained through a host computer and specifically includes: Prepare a training dataset; the training dataset includes original images of the production workshop and labeled images of the production workshop. The YOLOv4-tiny network is trained using the training dataset to obtain the trained YOLOv4-tiny network; Save the weight parameters and structure information of the trained YOLOv4-tiny network as a floating-point model file; The floating-point model file is quantized to obtain a fixed-point model file; The fixed-point model file is converted into the ZYNQ executable target detection file; The YOLOv4-tiny network comprises a backbone feature extraction network, a feature pyramid, and a YOLOhead output layer connected in sequence. The backbone feature extraction network is used to extract features from the production workshop image; the backbone feature extraction network includes an input layer, a first CBL convolutional layer, a second CBL convolutional layer, a first Resblock convolutional layer, a second Resblock convolutional layer, a third Resblock convolutional layer, and a third CBL convolutional layer connected in sequence; The feature pyramid is used to enhance the features to obtain enhanced features; the feature pyramid includes a first convolutional layer, a second convolutional layer, an upsampling layer, and a connection layer; the first convolutional layer is connected to the third CBL convolutional layer; the connection layer is connected to the third Resblock convolutional layer; The YOLOhead output layer is used to detect the enhanced features and obtain the detection results; The terminal device is used to display the detection results and issue an alarm based on the detection results.

2. The intelligent management and monitoring system for production workshops based on ZYNQ according to claim 1, characterized in that, The ZYNQ system board includes an SD card for storing the ZYNQ acceleration system and the ZYNQ executable target detection file; The ZYNQ acceleration system, located on the SD card, calls the ZYNQ executable target detection file; the ZYNQ acceleration system is also used to accelerate the detection process.

3. The intelligent management and monitoring system for production workshops based on ZYNQ according to claim 2, characterized in that, The ZYNQ system board further includes: a processing system terminal, a programmable logic terminal, a USB interface, an Ethernet interface, a display interface, and a serial port; the processing system terminal is connected to the programmable logic terminal; the processing system terminal is also connected to the USB interface, the Ethernet interface, the display interface, and the serial port respectively via an AXI communication bus; The processing system terminal is the main control terminal of the ZYNQ system board, used to provide clocks for the programmable logic terminal, the USB interface, the Ethernet interface, the display interface and the serial port; The programmable logic terminal is used to control the USB interface, the Ethernet interface, the display interface, and the serial port; The USB interface is used to connect the image acquisition module; The Ethernet interface is used to connect to the host computer; The display screen interface is used to connect to the display screen of the terminal device; The serial port is used for serial communication.

4. The intelligent management and monitoring system for production workshops based on ZYNQ according to claim 3, characterized in that, The terminal device includes a base, a display screen, and an alarm; The display screen and the alarm are connected; the display screen and the alarm are mounted on the base; The display screen is used to display the detection results, and the alarm is used to issue an alarm based on the detection results.

5. The intelligent management and monitoring system for production workshops based on ZYNQ according to claim 4, characterized in that, The ZYNQ system board is located inside the base.

6. The intelligent management and monitoring system for production workshops based on ZYNQ according to claim 3, characterized in that, The ZYNQ system board also includes a power on / off button.

7. The intelligent management and monitoring system for production workshops based on ZYNQ according to claim 1, characterized in that, The ZYNQ executable target detection file also includes the instruction set for the DPU-accelerated YOLOv4-tiny network and the main program for running ZYNQ.

Images