Accurate and rapid image segmentation method for bayberry automatic picking via machine learning

Due to the short ripening period and complex picking environment, bayberry generally relies on mechanical equipment for picking, especially the automatic picking system guided by vision. Thus, it is crucial to locate the bayberry in the view accurately and rapidly. Although efforts have been made, the existing methods are difficult to implement due to the limited amount of data and the processing speed. In this study, an accurate and rapid segmentation method based on machine learning was proposed to address this problem. First, the images collected by the visual guidance system were pre-processed by contrast-limited adaptive histogram equalization (CLAHE) based on the Y component of the YUV color space. Taking advantage of the color difference map of RB and RG for the segmentation of different colors, an adaptive color difference map foreground segmentation method was then adopted for bayberry region foreground segmentation. Finally, distance transforms and marking control watershed methods were exploited to achieve single bayberry fruit segmentation. Furthermore, with the help of the convex hull theory and fruit shape characteristics, the irregular background interference areas were filtered out, which improved the accuracy of bayberry segmentation performance. The experimental results show that this method can achieve better segmentation of bayberry in complex orchard environment with an accuracy of 97.4% and only takes 0.136 s to calculate once.