Super-Resolution Reconstruction of Crop Disease Images Based on Depth Learning

Abstract

Current image reconstruction has some problems, such as low image segmentation and denoising precision, slow convergence speed, and poor image integrity after reconstruction. In this regard, this study proposed a super-resolution reconstruction of crop disease images based on deep learning. The improved neighborhood averaging method is used to denoise the low frequency subband image, and the enhanced wavelet coefficients are replaced by the wavelet inverse transform to realize the high frequency subband image denoising. The image enhancement results are introduced, and the image initial segmentation area is obtained by using the color roughness concept and the incremental region growth method. When the stop condition is satisfied, the best result is presented. The small feature extraction of image segmentation results is carried out, and the feature vectors are transformed from low resolution space to high resolution space by nonlinear mapping. Feature extraction, nonlinear mapping and image reconstruction are fused into a deep convolution neural network, and the final reconstruction results are obtained. The experiment showed that the method improves the image segmentation and denoising precision, and that the integrity coefficient of the image reconstruction is high and the reliability is strong.