To achieve full-cycle monitoring of pomegranate growth under resource-constrained scenarios, this study proposes an improved lightweight algorithm YOLO-LFS, based on the YOLOv11 model. Firstly, ShuffleNetv2 is adopted to replace the backbone network of YOLOv11, and the MBConv module is introduced to replace the original detection head, reducing the computational complexity of the model and improving real-time detection capability. The Slide Loss function is incorporated to enhance spatial local information, increasing the focus on hard samples and thereby improving detection accuracy. Secondly, ablation experiments were conducted to compare the performance of different lightweight backbones and loss functions, validating the effectiveness of the modules selected in this study. Finally, the results of the controlled trial indicated that under the same conditions, YOLO-LFS can achieve efficient and accurate detection of pomegranate growth status while maintaining precision. This study provides a lightweight and high-accuracy solution for real-time monitoring and analysis of crops in agricultural production.