| With the rapid development of modern military technology,smart ammunition has become a research hotspot in the military field at this stage.Aiming at the shortcomings of the existing intelligent ammunition visual tracking system,such as few image pixels,large algorithm resource consumption,and low integration,an embedded image processing system based on Spike Neural Network(SNN)has been designed in this paper.The specific content is as follows:Firstly,according to the system requirements of the integration of acquisition and processing,an overall design scheme of embedded image processing system based on multicore ARM and FPGA is proposed,and through comparative analysis,the neuron model based on leaked integral firing(LIF)and the feedforward pulse neural network structure combined with the first pulse coding method provides a threshold-based SNN image segmentation processing scheme.On this basis,the selection of core components such as core processing boards and image sensors has been completed.Secondly,aiming at the disadvantages of traditional image segmentation algorithms of large amount of calculation and low precision,a simple structured feedforward impulse neural network was built,and an SNN image segmentation algorithm based on membrane potential threshold was designed.Aiming at the problem that the segmentation effect is difficult to evaluate,an evaluation function based on the two-dimensional entropy of the image is designed,and the comparison of the pros and cons of the segmentation results is achieved in a quantitative manner.The simulation experiment results show that the image segmentation algorithm based on SNN has a smaller amount of calculation and higher precision than traditional segmentation algorithms.Thirdly,in view of the problem that it is difficult for the commercially available image processing equipment to directly provide the raw data of the CMOS image sensor based on the SNN image segmentation,the software and hardware design of the embedded image processing system based on the pulse neural network was completed,including CMOS image sensor data processing circuit,the design of schematic diagram and PCB diagram,as well as the detailed design of software modules such as SPI communication,LVDS deserialization,image storage,and image segmentation.Finally,an embedded image processing system experimental platform based on pulse neural network was built,and the communication experiments,deserialization experiments,original image acquisition experiments,SNN-based image segmentation experiments were completed.The experimental results show that the two-dimensional entropy value of the image segmentation result based on SNN is infinitely close to 1,which confirms the effectiveness of the designed system. |
