A Lightweight Wheat Ear Counting Model in UAV Images Based on Improved YOLOv8

Ruofan Li ¹ Xiaohua Sun ^2* Kun Yang ^1* Zhenxue He ^1,3* Xinxin Wang ^4,5,6 Chao Wang ^1,3* Bin Wang ¹ Fushun Wang ^1,3* Hongquan Liu ^7*

• ¹ College of Information Science and Technology, Hebei Agricultural University, Baoding, China
• ² Department of Digital Media, Hebei software institute, Baoding, China
• ³ Hebei Key Laboratory of Agricultural Big Data, Baoding, China
• ⁴ Hebei Mountain Research Institute, Hebei Agricultural University, Baoding, China
• ⁵ Agricultural Technology Innovation Center in Mountainous Areas of Hebei Province,, Baoding, China
• ⁶ Agricultural Engineering Technology Research Center of National North Mountainous Area, Baoding, China
• ⁷ College of Urban and Rural Construction, Hebei Agricultural University, Baoding, China

Wheat (Triticum aestivum L.) is one of the significant food crops in the world, and the number of wheat ears serves as a critical indicator of wheat yield. Accurate quantification of wheat ear count s is crucial for effective scientific management of wheat fields. To address the challenges of misse d detections, false detections, and diminished detection accuracy arising from the dense distributi on, small size, and high overlap of wheat ears in Unmanned Aerial Vehicle (UAV) imagery, we p ropose a lightweight model, PSDS-YOLOv8 (P2-SPD-DySample-SCAM-YOLOv8), on the basis of the improved YOLOv8 framework, for the accurate detection of wheat ears in UAV images. Fi rst, the high resolution micro-scale detection layer (P2) is introduced to enhance the model's abilit y to recognize and localize small targets, while the large-scale detection layer (P5) is eliminated t o minimize computational redundancy. Then, the Spatial Pyramid Dilated Convolution (SPD-Con v) module is employed to improve the ability of the network to learn features, thereby enhancing t he representation of weak features of small targets and preventing information loss caused by low image resolution or small target sizes. Additionally, a lightweight dynamic upsampler, Dynamic S ample (DySample), is introduced to decrease computational complexity of the upsampling proces