Machine vision based guidance system for automatic rice transplanters

This article describes the development of a machine vision based guidance system for an automatic rice transplanter. The system is capable of analyzing images of the field shoreline (concrete or soil banks) and/or rows of seedlings, as well as having the functionality to detect the row and field end. Tests on more than 3000 images of concrete banks, soil banks, and rice seedlings showed that the average processing time was 3.27, 3.03, and 4.34 s, respectively. The detection accuracy was 99.2, 98.6, and 98.9% for concrete bank, soil bank, and rice seedlings, respectively, thereby verifying the applicability of the approach. The detection errors usually occurred when the mud was on the shoreline or near the rice line and when the seriate was missing seedlings. Avoiding these conditions was important to this system.