An infrared multi-object tracking method for automatic circadian behavior analysis of adult Tuta absoluta

• We propose an efficient multi-object tracking method, YOLO OCSORT, for automatic behavior analysis of adult Tuta absoluta . • The improved YOLO11 model is used for precise detection of adult Tuta absoluta . • The enhanced OCSORT algorithm is employed to achieve multi-object tracking of adult Tuta absoluta . • The performance of the YOLO OCSORT algorithm is analyzed under different adult Tuta absoluta density conditions. Adult Tuta absoluta tracking is a crucial prerequisite to automatically analyzing their circadian behavior in a cage-rearing environment. However, individual nonlinear motions and occlusions pose significant challenges to the effective tracking of multiple insects. In this study, an infrared multi-object tracking method (YOLO OCSORT) is proposed to mitigate these issues for circadian behavior analysis of adult Tuta absoluta . In the proposed approach, an Haar wavelet down-sampling module is integrated into the backbone of YOLO11 to enhance feature extraction under complex motion and occlusion conditions. Additionally, a multi-scale spatial and attention aggregation module is incorporated to preserve the detailed features of small objects by fusing spatial and channel information, while also enhancing the attention to target-relevant channels in the aggregation path. Lastly, a new detection head is added to the YOLO11 model to further boost detection performance of small adult Tuta absoluta . In the multi-object association stage, a GIoU-based association strategy is integrated into the OCSORT algorithm to improve the matching accuracy in complex scenarios such as occlusion and dense environments. Experimental results show that the proposed method achieves increases in all tracking metrics, with HOTA improving by 3.3%, MOTA by 8.9%, MOTP by 2.3%, and IDF1 by 3.0%, compared with the original model in the target tracking phase. Moreover, except a few false positives, no missed detections are observed under different density conditions. These results demonstrate that the proposed YOLO OCSORT method has robust performance in the multi-object tracking of adult Tuta absoluta , offering technical support for contactless behavior monitoring of small insects.