Unsupervised Learning Based Soil-film Recognition System for Electric Film Uncovering Machine

Abstract

To reduce the damage to the edge film of the soil ridge during the uncovering process of the film uncovering machine, the soil-film recognition system of the film uncovering machine is designed in this paper. Firstly, the overall structure and working principle of the electric film uncovering machine are analyzed. Then the aberration correction of the camera of the machine is performed. Subsequently, multiple processing methods for field soil-film images are compared and analyzed. The image segmentation method most suitable for this system is determined to be the K-means++ method. The acquired images are also filtered and noise is reduced, and the contour information of the film is finally extracted and identified. The range of feature values of the experimentally measured images is area 4500~20,000 mm2, area share 4%~22%, and perimeter 250~950mm. When the three conditions are satisfied at the same time, it is a soil-film image. The system is combined with the experimental film uncovering machine in the soil consolidation state. The average film recognition rate of the system is greater than 95%. At the same time, the damage rate of the film is below 5%, and the soil removal rate on the side film reaches 82%.