Research on Citrus Target Recognition Based on Improved YOLOv8 Algorithm

In natural orchard environments, tangerines are susceptible to being shaded by foliage and to overlapping with multiple fruits. Varying weather conditions can cause inconsistent levels of illumination, and these unstable factors combined with complex backgrounds can diminish the efficiency of tangerine recognition and localization. Consequently, this paper utilizes images of tangerines captured under various weather conditions within a tangerine orchard as a dataset, and a method based on the YOLOv8n object detection algorithm is proposed. The dataset was trained using BiFPN, MCA attention mechanism, and PConv. An improvement in the algorithm resulted in an accuracy rate of 94.4% for tangerine target detection, a recall rate of 92.7%, an F1 score of 93.5%, and a mAP of 98.3%, with each metric showing an increase of 0.7%, 0.6%, 0.7%, and 1.3% respectively over the original model.