AUTHOR=Zhang Mengli , Chen Wei , Gao Pan , Li Yongquan , Tan Fei , Zhang Yuan , Ruan Shiwei , Xing Peng , Guo Li 

TITLE=YOLO SSPD: a small target cotton boll detection model during the boll-spitting period based on space-to-depth convolution

JOURNAL=Frontiers in Plant Science

VOLUME=Volume 15 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2024.1409194

DOI=10.3389/fpls.2024.1409194

ISSN=1664-462X

ABSTRACT=Cotton yield estimation is an important part of the cotton production process, and the results of boll detection during the flocculation period directly determine the accuracy of yield estimation in cotton fields. Unmanned aerial vehicles (UAVs) are commonly used for plant detection and counting due to their low cost and flexibility. However, to address the problems of small target cotton bolls and low resolution of UAVs, this paper proposes a method based on the YOLO v8 framework of transfer learning, named YOLO small-scale pyramid depth-aware detection (SSPD). Based on YOLOv8, space-to-depth and non-strided convolution (SPD-Conv) and small target detection head are introduced. And a simple, parameter-free attention mechanism (SimAM) is added. The results show that YOLO SSPD has a boll detection accuracy of 0.874 at the UAV scale, and the coefficient of determination (R² ), root mean square error (RMSE), and relative root mean square error (RRMSE) results for boll counts are 0.86, 12.38 and 11.19%, respectively. This study suggests that YOLO SSPD can improve the accuracy of cotton boll detection on UAV imagery to support the cotton production process.