Research on Detection Technology of Various Fruit Disease Spots Based on Mask R-CNN

Abstract

In order to solve the current fruit surface disease detection algorithm's problems of low accuracy, slow speed and heavy workload of quality classification, this paper takes apple, peach, orange, and pear as the research objects and proposes a model based on Mask R-CNN for detecting disease spots on the surface of fruits which accurately detects the defects on the surface of the fruit after the picking robot recognizes and locates the fruit. By adding a bottom-up horizontal connection path, the feature pyramid (FPN) structure of Mask R-CNN is improved to enhance the fusion of high and low-level features. Experimental research shows that the improved Mask R-CNN algorithm has a detection accuracy of more than 95% for the four kinds of fruit surface lesions, and the detection speed reaches 2.6 frames per second when using GPU, which is significantly better than Fast R-CNN and SSD algorithms and has good detection performance and robustness.