Apple appearance quality classification method based on double branch feature fusion network

This paper carried out a classification study on the quality of different types apples. In order to achieve high-precision high-quality apples in various dimensions, shorten time, improve efficiency, and make it feasible for practical applications. proposes an automatic recognition model OB-Net (for apple appearance quality) based dual-branch structure. The consists two branches, O branch B branch. extracts shape contour features surface defect by decomposing feature map into high low frequencies. size texture characteristics fusing channel attention spatial mechanisms; this is done further increase distance between defects. Experimental results show that accuracy rates rot, insect bites, russeting, scratches, intact are 95.65%, 98.17%, 94.62%, 92.02% 97.67%, respectively; overall rate reaches 95.64%. Finally, three aspects map, heat category probability statistics extracted from network prove validity quality.