Parallel Optimization Of Immune Covolution Nerual Network And Its Application With Embedded System

Nowadays, image recognition technologies have been widely used in many fields, includingunmanned vehicles, intelligent monitoring and even outer space exploring. Convolution neuralnetwork (CNN) is a technology that combines the artificial neural network (ANN) and the deeplearning method. It is characterized by local receptive filed, hierarchical structure, global learningfor feature extraction and classification and has been applied to many image recognition fields.Evolutionary computation, artificial neural networks and artificial immune system have manythings in common, but also have their own characteristics. Adaptation, learning and memorizingability of the immune system can not only be applied to the field of optimization, but alsomachine learning, such as artificial immune system. Although there are a lot of learning theoriesand methods, none is perfect and omnipotent. They just complement each other.In this work, the development of artificial neural networks, convolution neural network,theembedded image processing system and GPU accelerated parallel processing and the latestresearch results at home and abroad were reviewed and summarized. The concepts and algorithmsof artificial neural networks and classical convolution neural network were briefly introduced.Finally, an embedded real-time image recognition system was established, which adopted thetheories of artificial immune system and convolution neural network,called I-CNN. The algorithmwas implemented on GPUs, which efficiently accelerated its speed.The main work is designed as follows:1. Analyze the network structure and parameters of the convolution neural network, thecomplex network structure, the long time training, the over fitting and classification accuracy isnot higher,in order to overcome this difficulty, A immune convolution neural network algorithmbased on the advantages of artificial immune algorithm was established, The proposed algorithmmerges the positioning and parameter adjustment of network nodes as well as the parameter tuningof basis functions. 2. Use the cuDNN depth neural network library from NVIDIA to accelerate the process formachine learning, so that deep neural network model can be used for real-time convolution inembedded platforms.3. Complete the setting up of image recognition system based on ARM+Linux. Thisincludes the tailoring of Linux operating system, the driver module of hardware, and thepre-processing of the signals from the sensors,The parallel optimization of immune convolutionneural network and the application in embedded image recognition system.