RS-Net: robust segmentation of green overlapped apples

Fruit detection and segmentation will be essential for future agronomic management, with applications in yield estimation, growth monitoring, intelligent picking, disease etc. In order to more accurately efficiently realize the recognition of apples natural orchards, a robust net framework specially developed fruit production is proposed. This model was improved challenging problem which segments overlapped from monochromatic background regardless various corruptions. The method extends Mask R-CNN by embedding an attention mechanism focusing on informative pixels but also suppressing noise caused adverse factors (occlusions, overlaps, etc.), could suitable operating complex environment. Specifically, Gaussian non-local transplanted into refining semantic features generated continuously residual network feature pyramid network, then forward processing based balanced levels finally regions where are located. Experimental results verify hypothesis current work show that proposed outperforms other start-of-the-art models, AP box mask metric values have reached 85.6% 86.2% reasonable run time, respectively, can meet precision robustness vision system management.