Improved Lightweight YOLOv8-Based Detection of Surface Defects in Wokan Oranges

Wokan oranges are highly favored in the market due to their large yield, sweet taste, and rich vitamin content. However, Wokan oranges are susceptible to surface pests and physical damage during growth and storage, leading to fruit decay, economic losses, and a poor first impression for consumers. This makes surface defect detection for Wokan oranges a research priority. Currently, defect removal relies mainly on manual inspection, which is time-consuming and labor-intensive. To address this, this paper proposes an improved lightweight detection method based on the newly released YOLOv8n model to automatically identify and screen surface defects in Wokan oranges. The traditional Conv module is replaced with the GhostConv module in the backbone network, and the MDSA attention mechanism is introduced in the detection head. The modified YOLOv8-GhostConv-MDSA model outperforms the original YOLOv8n with a 2.0% increase in recall, 6.2% in precision, and 5.6% in [email protected]. It reduces model parameters by 16.8% and GFLOPs by 8%, improving accuracy, speed, and efficiency for deployment on low-power platforms. This lays a solid foundation for future fruit defect detection and grading.