基于不同目标检测模型的胡椒园环境下胡椒果穗成熟度判别研究

Research on the Discrimination of Pepper Fruit Spikes Maturity in Pepper Orchard Environment Based on Different Object Detection Models

目前,针对胡椒智能采摘领域的研究尚属空白,精确识别胡椒果穗的成熟度是胡椒智能采摘关键技术之一。通过两条技术路线判别胡椒成熟度:一是先建立胡椒目标检测深度学习模型,再根据胡椒果穗颜色特征来判别胡椒成熟度;二是直接建立胡椒成熟度判别深度学习模型。利用两条技术路线,采用SSD、Faster R-CNN、YOLOv5s、YOLOv5m和YOLOv8m这5种算法进行胡椒果穗成熟度判别对比。研究结果发现,基于YOLOv8m模型,第1种方法成熟度判别的准确度为94.81%,第2种方法的准确度、召回率等多项指标高达98%以上,可为胡椒智能化采摘机器人的开发提供依据。

Currently,research in the field of smart pepper harvesting remains unexplored,and one of the key technologies involves accurately identifying the maturity of pepper fruit spikes.This study employed two technical routes to determine pepper maturity.The first method was to establish a deep learning model for pepper target detection,followed by assessing pepper maturity based on the color characteristics of the pepper fruit spikes.The second method was to directly establish a deep learning model for determining pepper maturity.Both methods utilized five algorithms for comparison,including SSD,Faster R-CNN,YOLOv5s,YOLOv5m and YOLOv8m,to discriminate pepper fruit spikes maturity.The research results found that based on the YOLOv8m model,the first method achieved a maturity discrimination precision of 94.81%,while the second method demonstrated an excellent performance,with precision,recall rate and other metrics exceeding 98%,providing an important basis for the development of intelligent pepper harvesting robots.