Yolov8-CGD: An Improved Model for Lemon Maturity and Quality Detection, Suitable for Complex Field Environments

To achieve precise detection of lemon maturity and quality in complex field environments, this study proposes a high-accuracy model for lemon maturity and quality detection in such environments (YOLOv8-CGD). In order to improve the model’s accuracy in capturing lemon images in complex environments, the ConvNeXt V2 module is introduced into the backbone network. Subsequently, the Gather-Excite attention mechanism is introduced to utilize spatial context information, further enhancing the model’s ability to capture key features. The DIoU loss function is used instead of the CIoU loss function to improve the model’s detection accuracy and positioning stability for overlapping and small targets in complex environments. The results show that the YOLOv8-CGD model achieves an accuracy of 94.4%, a recall rate of 92.8%, mAP50 and F1 scores of 97.2% and 93.6%, respectively, representing improvements of 4.9%, 4.3%, 3.2%, and 4.6% compared to YOLOv8. Ablation experiments further validate the effectiveness of the improved model, and heatmap analysis also demonstrates that the YOLOv8-CGD model captures features more precisely. Comparison results between the YOLOv8-CGD model and other mainstream models such as YOLOv11s, YOLOv10s, YOLOv7-tiny, and YOLOv5n show that the YOLOv8-CGD model is superior in accuracy, mAP50, and F1 score, demonstrating excellent overall performance. A frame rate of 237.3 FPS also meets the real-time monitoring requirements. This study provides an efficient and reliable method for lemon maturity and quality detection, offering technical support for high-precision automated harvesting in robotic pickers.