A Lightweight Algorithm for Detection and Grading of Olive Ripeness Based on Improved YOLOv11n

Olives are a crucial woody oil crop, the harvesting of which predominantly relies on manual labor, which is characterized by high costs, low efficiency, and challenges in ensuring optimal harvesting timing. The development of an automated ripeness-detection system with high recognition accuracy is of paramount importance for advancing automated olive-harvesting technologies. Traditional detection methods are constrained by susceptibility to environmental interference, limited robustness, and inadequate generalization capabilities. Concurrently, existing deep learning-based detection models face issues such as insufficient feature extraction for small targets and difficulties in deployment due to their need for large numbers of parameters. To address these limitations, this study proposes a lightweight algorithm for detection and grading of olive ripeness based on an Improved YOLOv11n framework. The proposed approach employs YOLOv11n as the baseline model, replaces its backbone network with EfficientNet-B0, and integrates the Large-Scale Kernel Attention (LSKA) mechanism and the Bidirectional Feature Pyramid Network (BiFPN). Experimental validation demonstrated that the enhanced model achieved detection accuracy comparable to that of the original model, attaining a mean average precision (mAP) of 0.918. Furthermore, the model size was reduced to 3.7 MB, a 39.3% reduction, while the computational complexity (GFLOPs) was decreased by 2.4 and the inference time per image was reduced by 0.2 ms. The proposed model exhibits significant advantages in terms of lightweight design and improved detection efficiency, demonstrating substantial potential for practical deployment. This study provides a valuable reference for the development of automated olive-harvesting technologies.