<p>Lightweight object detection algorithms that maintain a balance between speed and accuracy are essential for strawberry-picking robots. Strawberry fruit detection, however, is often challenged by occlusions from stems and leaves, overlapping fruits, and complex background interference. In addition, most existing ripeness detection methods rely on high-performance computing platforms, which limits their practicality in field applications. To address these issues, a lightweight real-time strawberry ripeness detection algorithm, SR-YOLO, is proposed based on the YOLO11 framework. A compact module, termed the Pela block, is designed using partial convolution (PConv) to reduce parameter count and memory access, while simultaneously suppressing background interference. A lightweight localization-based asymptotic feature pyramid network (LAFPN) is incorporated to enhance multi-scale detection performance. Moreover, a novel Focaler-PIoU loss function is introduced to direct the model toward anchor boxes of moderate quality, thereby mitigating the difficulty imbalance among strawberry samples. A large-scale, complex dataset is constructed, comprising 14,300 images and over 65,000 annotated strawberry instances, to evaluate the model. Experimental results indicate that the enhanced model achieves a precision of 91.3% and an mAP50 of 92.1%, with improvements of 2.8% and 2.5% over the baseline, respectively. Compared to mainstream detection algorithms, the proposed model demonstrates superior accuracy while maintaining a reduced number of parameters and lower computational complexity. When deployed on a Jetson AGX Orin embedded edge device with FP16 quantization, it sustains a detection speed of 90.1 FPS. In conclusion, SR-YOLO offers accurate, real-time detection of strawberry ripeness across complex agricultural environments, providing practical value for the vision modules of autonomous strawberry-picking robots.</p>