Abstract: Accurate and rapid detection of grape leaf diseases is crucial for ensuring grape industry healthy development．Aiming at detection challenges in complex field environments, such as small disease targets, variable morphologies, and strong background interference, an target detection model based on improved YOLOv8 has been proposed．Firstly, an adaptive spatial correlation pyramid attention（ASCPA）mechanism has been introduced, which significantly enhanced model's ability to capture details and distinguish features of small-sized lesions through multi-scale feature fusion and adaptive weight allocation．Secondly, Wise-IOU loss function has been employed to replace traditional bounding box regression loss．By dynamically focusing mechanism on gradient updates for different qualities' samples, this approach has effectively improved lesions' localization accuracy, especially in overlapping and edge-blurred areas．Finally, a warm-up cosine annealing scheduler has been used to adjust learning rate and batch collaborative optimization strategy, effectively preventing overfitting and significantly accelerating training convergence, while introducing complex modules to increase model capacity．Experimental were conducted on a self-built grape leaf disease dataset, and results have shown that improved model achieved precision, recall, and mean average precision（mAP50）of 0.99228, 0.99345, and 0.99384, respectively, representing significant improvements over baseline model．Furthermore, training efficiency has been increased by approximately 38%, which achieved simultaneous optimization of accuracy and efficiency, providing an effective solution for intelligent disease detection in field．