Recognition of Tightly-Stacked Bagged Agricultural Materials for a Handling Robot

Abstract

Aiming at the situation that the material bags in the train compartment are stacked tightly, a new marking method for watershed algorithm is proposed. After sufficient experiments and comparison, it can be used in image segmentation under multi-material bag stacking scenes. Firstly, a grayscale morphology filter was used to smooth the image. Secondly, morphological reconstruction was applied to find the foreground and background markers. Thirdly, local entropy of the image was used to generate the initial texture gradient image, and the final gradient magnitude was obtained through minima imposition using the initial texture gradient and the markers. Finally, watershed segmentation algorithm was applied on the final gradient magnitude and achieved the effective segmentation of the connected bags. The segmentation accuracy was 94%, which is higher than marker-controlled watershed based on Sobel operator.