In intelligent multi-agent systems, particularly in drone combat scenarios, the challenges posed by rapidly changing environments and incomplete information significantly hinder effective strategy optimization. Traditional multi-agent reinforcement learning (MARL) approaches often encounter difficulties in adapting to the dynamic nature of adversarial environments, especially when enemy strategies are subject to continuous evolution, complicating agents’ ability to respond effectively. To address these challenges, this paper introduces a novel enhanced MARL framework, MADDPG-SASP, which integrates an improved self-attention mechanism with self-play within the MADDPG algorithm, thereby facilitating superior strategy optimization. The self-attention mechanism empowers agents to adaptively extract critical environmental features, thereby enhancing both the speed and accuracy of perception and decision-making processes. Concurrently, the adaptive self-battling mechanism iteratively refines agent strategies through continuous adversarial interactions, thereby bolstering the stability and flexibility of their responses. Empirical results indicate that after 600 rounds, the win rate of agents employing this framework saw a substantial increase, rising from 26.17% with the original MADDPG to a perfect 100%. Further validation through comparative experiments underscores the method’s efficacy, demonstrating considerable advantages in strategy optimization and agent performance in complex, dynamic environments. Moreover, in the Predator–Prey Scenario combat environment, when the enemy side employs a multi-agent strategy, the win rate for the drone agent side can reach 98.5% and 100%.

Pest and disease detection in natural pear orchard environments is often affected by branch and leaf occlusion, non-uniform illumination, and complex background textures. In addition, substantial variations exist in the morphology and scale of lesions and insect-damaged regions, while the weak texture features of early-stage lesions can be progressively attenuated during feature propagation. Therefore, effectively representing targets at different scales while preserving discriminative local details and suppressing irrelevant background information is crucial for improving pest and disease detection in pear orchards. To address these challenges, this study proposes BAP-YOLO, a detection framework designed for natural pear orchard environments. The framework progressively enhances feature representation through context aggregation, adaptive scale modeling, and discriminative detail refinement. First, a Branch-based Target Enhancement Fusion (BTEF) module is introduced into the backbone network to enhance hierarchical feature representations at multiple backbone stages, thereby reducing interference from complex background information in target representation. Second, an Adaptive Scale Dynamic (ASD) module is designed and embedded into the neck network to construct candidate features with different effective receptive fields and dynamically modulate their contributions using multi-resolution statistics, thereby improving the model’s ability to detect targets with substantial scale variations. Finally, a Parallel Nonlinear Enhancement (PNE) module is incorporated between the neck network and detection head to enrich weak-texture and local-detail representations through complementary nonlinear transformations and element-wise fusion before prediction. Experiments were conducted on a pear orchard image dataset containing six classes: beetle, black_block, dry blight, health, rust, and scab. The experimental results showed that BAP-YOLO achieved a precision of 0.854, an mAP@50 of 0.876, and an F1-score of 0.826. Compared with the baseline YOLO11n, BAP-YOLO improved Precision, mAP@50, and F1-score by 6.1, 2.1, and 1.3 percentage points, respectively. These results demonstrate the potential of BAP-YOLO for pear disease and pest detection in natural orchard images.