Design Of Optical Test Platform For Intelligent Visual Perception Chip

With the growth and maturity of science and technology,the current machine vision can completely replace us to interact with the outside world.Intelligent visual perception,on the one hand,can greatly improve the quality of data collection,reduce the algorithm consumption of back-end data processing,and reduce the overall cost;on the other hand,the data processing method based on the layered signal processing architecture also improves the reliability of the system.In view of the current development of image recognition technology is quite mature and applied in all aspects of life,this paper takes image recognition as the target task,designs and builds an agile test optical experimental platform for intelligent visual perception chips.The main research contents are as follows:(1)The hardware circuit design of the experimental platform.The experimental platform mainly completes the intelligent display of images to be classified,data collection and classification,and classification calculation results display.Based on this,the optical display circuit of the image is designed with the LED driver chip SM16126N-2 and the driver amplifier 74HC245 as the core;the always-on smart CMOS image sensing chip is used to convert the optical signal of the image into an electrical signal;the FPGA chip XC7Z020-1CLG484 C is used as the core of the development module of the processor controls the normal operation of each functional circuit,and finally connects to the PC through VGA and serial port to display the collected original images and classification results.(2)Use binary neural network and convolutional neural network to realize image recognition and classification.This paper presents the principles of the two algorithms of Binary Neural Network(BNN)and Full Precision Convolutional Neural Network(CNN)and the implementation of the application in this paper.The 10,000 handwritten digits "0-9" in the MNIST data set are used as the identification target and the data set is collected,and the algorithm model is built and the model is trained.(3)Build an optical darkroom.This paper designs an optical darkroom,in which the optical vibration isolation table is used to reduce the impact of environmental vibration,the lens hood is used to avoid the impact of ambient light,the precision sliding table and the manual lifting sliding table are used to adjust the position of the PCB board,and it is also designed the CMOS image sensor circuit board and the mechanism fixing device of the LED display screen.Based on this,the experimental system can be carried out in a completely opaque operating space,effectively preventing the influence of light interference and environmental vibration in the experimental environment,and maintaining the stability of the optical path.Based on the above research content,complete the performance test and evaluation of the intelligent visual perception chip,including linear compensation,classification accuracy and power consumption.After testing,when the chip's power supply voltage is0.8V and the exposure time is 400 us,the performance of the experimental system is the best.Its classification accuracy can reach 93.76%,the chip power consumption is147.84 n W,and the frame rate is 156 fps.At the same time,it took 35 minutes to display5011 sample pictures on the LED screen.This shows that the design has good image recognition and classification capabilities,and the entire experimental platform has a high degree of intelligent verification and operating efficiency.