Spectral Image Recognition Based on Convolution Neural Network

Spectral Image Recognition Based on Convolution Neural Network

연구 분야: Artificial Intelligence

학회: 2022 5th International Conference on Pattern Recognition and Artificial Intelligence (PRAI)

In traditional image target recognition technology, the distribution angle of the main direction of the target feature points is large, which leads to poor target recognition accuracy. Therefore, a target recognition technology based on CNN is proposed. The spectral image is transformed into gray image, and the image is enhanced. The linear mean filter is selected to remove impulse noise and Gaussian noise. The Hessian matrix template is applied to the gray image. Each pixel in the rectangular region is compared to obtain the local extremum. The spatial scale of the target contour and edge is preliminarily determined. Then, the target is convoluted layer by layer, and the iterative training is completed The detailed features are extracted and matched with the template image features to identify the target category. The robot grabs the visual spectrum image of the aircraft target, and carries out the comparative experiment. The results show that, on the basis of ensuring the recognition time, this design technology is closer to the template feature quantity than the traditional technology, and improves the recognition accuracy of the image target.