| The integration of real-time detection and recognition of targets has currently been a hot research topic in hardware recognition technology,as well as an effective means to achieve the efficiency and intelligence of hardware recognition.Its main application is to use network frameworks and related algorithms to complete the design.However,due to the high complexity of the network structure,high resource consumption,and long calculation cycle,more and more researchers are trying to complete algorithm design and transplantation on integrated chips.This paper focuses on the recognition of specific targets in complex environments and conducts research on infrared image acquisition,typical target recognition algorithms,and hardware design implementation to solve the problems of all-weather target recognition and display.In the hardware design and implementation process,there are often difficulties in data interaction and high complexity in the collaborative development of multiple embedded platforms.To solve this problem,this paper will conduct research and design based on the hardware implementation of image acquisition and display and target recognition algorithms.As a result,it is necessary to optimize the design of infrared image acquisition and display,as well as target recognition algorithms,to complete the development of a hardware system for typical infrared target recognition.It mainly consists of three parts: the infrared image acquisition and control system,the hardware implementation of typical infrared target recognition algorithm,and the hardware system construction and testing.The main contents can be listed as following:1.Design of infrared image acquisition and control system: Based on the overall hardware system scheme of the infrared image acquisition and control system,the selection of infrared detectors,main control chips,and storage devices is completed.According to the analysis and calculation of various data parameters by the hardware core components,and the work principles of the control module,image acquisition module,cache module,and display module,the design and functional simulation of data storage,data read and write,and output modules are completed to verify the correctness and rationality of the system design.2.Completion of typical infrared target recognition algorithm and hardware implementation:Firstly,the target images of tanks and armored vehicles are collected and preprocessed to establish a sample dataset.Secondly,the working principles of the convolutional neural network model and each network layer are analyzed,and the convolutional neural network is optimized to complete the design of the target recognition algorithm.Finally,the algorithm is implemented in hardware layout on the main control.In the programmable logic end of the main control chip,the convolution operation is optimized and designed to fully realize the advantages of parallel acceleration calculation.This process is mainly to complete the design and simulation verification of modules such as matrix sliding window,convolution multiplication,addition,activation function,and pooling operation.The modules with many quantities and small calculation amount,such as image storage and weight file reading,are implemented in the embedded control end of the main control to complete the design of network parameter transmission,operation data caching process,and fully connected layer.3.Construction of typical infrared target recognition hardware system: The infrared image acquisition and recognition systems are constructed.In terms of software,the software design process of each module is introduced in detail,and the executable file of the software project is generated.In terms of hardware,the module design of the typical infrared target recognition hardware system is completed,and its algorithm function is hardware verified.Through experimental testing,the infrared image acquisition and control system built can realize the acquisition,storage,and display of infrared images with a resolution of 1280*1024under the condition of a clock frequency of 50 MHz.The developed typical infrared target recognition hardware system can recognize tanks and armored vehicles in different scenarios and different occlusion situations.The hardware system has low power consumption,reasonable resource utilization,and certain application value. |
