Just Accepted

This article has been peer reviewed and accepted for publication. It is in production and has not been edited, so may differ from the final published form.

Research on the method of rice panicle detection based on improved YOLOv5

Xiaoyue Seng, Xue Yang, Tonghai Liu , Rui Zhang, Chuangchuang Yuan, Tiantian Guo, Wenzheng Liu

Abstract

Context: Rice panicle counting is the key to agricultural production. Studying it is helpful to improve efficiency, optimize resources, promote variety breeding and farmland management. Aims: In order to count rice panicles conveniently and quickly, a rice panicle recognition model based on YOLOv5s-Slim Neck-GhostNet was proposed. Methods: In this paper, the stage from heading stage to maturity stage of rice is taken as the research object. On the basis of YOLOv5s model, GSConv convolution module is used to replace the original ordinary Conv convolution. In addition, we also improve the original C3 module and replace it with VoVGSCSP module, which further enhances the detection ability of the model for small targets. In order to further optimize the performance of the model and reduce the computational complexity, we replace the backbone network of the model. Specifically, we replace the original backbone network with a lightweight and efficient GhostNet structure. Key results: The experimental results show that the precision of the method on the test set is 96.5%, the recall rate is 94.6%, the F1-score is 95.5%, and the mAP@0.5 is 97.2%. Compared with the original YOLOv5s model mAP@0.5, it is increased by 1.8%, and the model size is reduced by 5.7M. Conclusions: The results show that the method reduces the size of the model while ensuring the accuracy, and can effectively detect and count rice panicles. Implications: This study can provide important data support for scientific research and agricultural intelligence, and promotes the process of agricultural modernization.

CP24073  Accepted 20 January 2025

© CSIRO 2025