Research On Cifar-10 Image Recognition System Based On FPGA

With the advancement of science and technology and mathematical methods and the emergence of various excellent algorithmic frameworks,the ability of computers to deal with problems has become stronger and there are more and more types of problems.Artificial intelligence has gradually become a solution that can solve practical problems in daily life.field of.In the past,humans were able to easily recognize objects through continuous learning,and computers also needed a lot of learning to recognize some easily distinguishable objects.Artificial intelligence is based on the neural network of the human brain,through a series of mathematical models and rules,to build a layered network to classify and recognize images and sounds.Therefore,the reason why artificial intelligence can quickly classify and identify is based on the need for a large amount of data for the previous training model,with powerful computing power hardware equipment and a more excellent algorithm framework,these three are indispensable.This article starts with image transmission and display,establishes a mathematical model of the image transmission and system,uses Modelsim and Matlab software to perform corresponding simulation tests on the mathematical model,builds an image transmission and display system,and verifies the practicability of the image transmission and display system.Then learn the concept of deep convolutional neural network in depth,understand the composition of the deep convolutional neural network model and the corresponding role of each component,and then build the deep convolutional neural network model to train the corresponding data set and test the correctness of the model According to the correct rate of the model,adjust the basic rate of the model to make the model have the best recognition rate,extract the tested model parameters through high-level programming language,design the corresponding image recognition module,and then build the overall image recognition system.Observe the hardware resources occupied by internal modules through the observation window of the SoC platform,and perform data analysis on the modules occupying more resources,so as to perform corresponding acceleration operations on the system modules,so as to optimize the overall structure of the system.The image recognition system designed in this paper tests the Cifar-10 data set and trains the model 100,000 times.The theoretical correct rate of the test can reach about 0.7639.The experimental results show that the image recognition system has only a few recognition errors,and the rest are recognized correctly.Basically meet the initial design requirements.The image recognition system designed in this paper still has a good recognition rate under bad lighting conditions,so the system has good advantages in robustness and adaptability.