Introduction Manual detection of peanut leaf diseases is plagued by a significant time lag, which frequently enables diseases to develop from isolated, sporadic outbreaks into large-scale epidemics. This delay ultimately leads to substantial regional yield losses in peanut production. Consequently, the precise detection capability of intelligent monitoring equipment is essential for mitigating the risk of large-scale peanut disease outbreaks. Detection algorithms serve as the core technology underpinning intelligent detection devices, highlighting the need for optimized, high-performance algorithms to address this challenge. Methods This study takes the YOLOv12 algorithm as the baseline model and proposes an improved model named YOLO-SDA. To enhance the model’s performance while reducing its computational burden, three key modules—StarNet, DySample, and A2C2fSCSA—are integrated into the original YOLOv12 framework. The integration of these modules is designed to optimize feature extraction, sampling efficiency, and feature fusion, thereby improving the model’s detection accuracy and reducing its resource consumption. Results Experimental results demonstrate that the proposed YOLO-SDA model outperforms the baseline YOLOv12 model in both performance and efficiency. Specifically, compared with YOLOv12, the YOLO-SDA model achieves a 44% reduction in parameters, a 38. 5% decrease in GFLOPs (giga floating-point operations per second), and a 43. 6% reduction in model size. Simultaneously, the model’s detection precision and mAP@0. 5–0. 95 (mean average precision at intersection over union thresholds from 0. 5 to 0. 95) are improved by 2. 0% and 2. 5%, respectively. Discussion The superior performance of the YOLO-SDA model confirms the effectiveness of integrating StarNet, DySample, and A2C2fSCSA modules into the YOLOv12 framework. The significant reduction in parameters, GFLOPs, and model size addresses the practical challenge of deploying intelligent detection algorithms on resource-constrained equipment, making it more suitable for on-site peanut leaf disease monitoring. The concurrent improvement in detection precision and mAP@0. 5–0. 95 ensures that the model can accurately identify peanut leaf diseases even in complex field environments, providing a reliable technical support for preventing large-scale disease outbreaks and safeguarding peanut yield.
Yang et al. (Tue,) studied this question.