Memristor-based RDBO-CNN circuit design and application of image multi-classification recognition.

Traditional convolutional neural networks used for classification largely rely on hyperparameter tuning and do not have the conditions for hardware implementation. Therefore, a memristor crossbar architecture circuit is proposed to implement the reinforced dung beetle optimization (RDBO) algorithm and the convolutional neural network (CNN). The circuit is composed of feeding module, storage module, ball rolling module, dance module, subpopulation module and CNN module. Traditional DBO algorithm with its adaptability and parallelism for CNN parameter optimization, there are some shortcomings. To solve the problem of unbalanced exploration and exploitation, the tendency to fall into local optimal state, an enhanced dung beetle optimization algorithm based on giant dung beetle and spiral search is proposed. The RDBO circuit is composed of feeding module, storage module, ball rolling module, dance module and subpopulation module. The CNN module is composed of convolution layer, pooling layer and fully connected layer, which is used to recognize and classify the image. The feasibility and accuracy of RDBO-CNN circuit are verified on MNIST image set. In order to further verify the effectiveness of the proposed circuit, simulation and comparison experiments are carried out the satellite image recognition RSI-CB image set which also has good accuracy. This will further promote the development and application of neural network technology.