Wheat ear detection based on FasterCANet-YOLOv8s algorithm

ABSTRACT

The application of neural networks in wheat ear detection and counting in smart agriculture holds significant value, showcasing how artificial intelligence technology brings innovation and improvement to the agricultural sector. However, due to the high density, variety, and complex growth cycles of wheat ears, as well as the influence of complex backgrounds during detection, issues such as false positives, false negatives, and low detection rates may arise. Addressing these challenges, this paper proposes a wheat ear detection solution based on the FasterCANet-YOLOv8s algorithm. Firstly, the FasterCANet Block is introduced to enhance the speed of wheat detection. Secondly, an efficient network structure, the QAFPN model, is proposed to strengthen the Neck network, achieving a balance between speed and accuracy in wheat detection. Finally, to better capture the features of small targets, the RFB Block is introduced to improve the SPPF layer. The improved algorithm achieves an [email protected] of 94.4%, surpassing existing technologies. Our model can accurately and quickly locate targets, holding tremendous potential in the wheat cultivation domain and providing efficient and precise support for agricultural production.

KEYWORDS:

• Smart agriculture
• deep learning
• wheat ear
• YOLOv8s
• feature fusion
• attention mechanism

Disclosure statement

No potential conflict of interest was reported by the author(s).