Recognition of incomplete and overlapped weed seeds based on self-organization delayed neural network

Abstract

The polygonal representation method is put forward in the paper. The method is based on recursion and boundary division which describes the shape of the incomplete and overlapped weed seeds. The method extracts the contour shape features as local features using the scale space method. The local features are irrelevant to the position and orientation, at the same time, meet the scale, rotation and translation invariance. The incomplete and overlapped weed seeds are identified using the self-organization delayed neural network. The adjacent corner features are analyzed, compared and identified by spatial adjacency relationship among the angle characteristics. At last, the method was proved feasible the experiment by in recognizing the incomplete and overlapped weed seeds.