This study addresses the challenges of irregular mulberry fruit distribution and complex background interference in natural environments, which often result in missed and repeated detections, by proposing an optimized mulberry maturity detection method based on an improved YOLOv8n model. A notable innovation involves integrating the SimAM attention mechanism after the SPPF module in the backbone network, enhancing mulberry feature extraction through an energy function while suppressing background noise without additional parameters, thus improving computational efficiency. Additionally, the C2f module is replaced with the C2f_Faster module, reducing computational complexity and parameters to enhance inference speed, representing a novel lightweight design. Furthermore, the CIoU loss function is substituted with Wise-IoU to mitigate repeated detections, enhance fruit localization precision, and reduce early training gradient oscillations, offering a robust optimization approach. Experimental results reveal that the enhanced YOLOv8n model achieves a 2.6% increase in mAP 0.5 over the original model, with a 23.4% reduction in parameters and a 1.5MB decrease in model weight, demonstrating significant lightweight improvements. These findings highlight the model’s superior accuracy and Lightweight deployment, providing a promising framework for deployment on picking robots and hardware systems.