Detection and Counting of Sesame Capsules per Plant Based on Improved YOLOv8-Track Model

LI Chenhao; WANG Chuan; LI Guoqiang; ZHAO Qiaoli; YANG Ping; WANG Kai; CHANG Shenglong; ZHENG Guoqing

doi:10.15933/j.cnki.1004-3268.2025.04.016

Abstract：Sesame capsules are an essential factor in the composition of sesame yield.In order to realize the accurate detection and counting of sesame capsules per plant,using object detection,multiple targets tracking and other technologies for dynamic tracking of capsules per plant is helpful to improve the efficiency of sesame breeding and cultivation management.Aiming at the phenomena of sesame capsules,such as small target,dense growth and overlapping occlusion,this study taked YOLOv8-Track as the benchmark model,introduced small target detection head and Shuffle attention mechanism into the feature fusion network,and introduced MPDIOU loss function in the post-processing stage of the model to construct SD-YOLOv8-Track model.In addition,this study utilized the ID counting method of model ByteTrack multi-target tracking algorithm to track and count sesame capsules using a single rotating video of sesame as the model input.The results showed that when taking a single picture as input,the accuracy,recall,and mean average precision of the SD-YOLOv8-Track model for detecting capsules were 92.25%,92.4%,and 92.58%,respectively,indicating 5.94,6.6,and 6.31 percentage points higher than those of the original model YOLOv8-Track.For the rotating video as input,the multiple object tracking accuracy and multiple object tracking precision of SD-YOLOv8-Track model were 89.42%and 88.23%,respectively,which were 4.23 and 4.60 percentage points higher than the original model.The accuracy,missed detection rate,and error detection rate of the SD-YOLOv8-Track model were 93.27%,3.85%,and 2.88%,respectively.The accuracy rate was 5.61 percentage points higher than that of the original model,and the missed detection rate and false detection rate were 3.84 and 1.77 percentage points lower than that of the original model.The improved SD-YOLOv8-Track model performs better in detecting sesame capsules and is suitable for dynamic complete counting of sesame capsules in a plant.