Abstract: In aquaculture, the traditional identification and diagnosis of parasitic infections rely heavily on manual microscopic examination by professionals’ experiences. However, this approach is time-consuming and subjective, often leading to inaccurate results due to human error. Furthermore, the diverse morphology of parasites, including variations in size, shape, and position, as well as issues such as occlusion and overlapping, all of which contribute to high false positives and missed detections rates in manual inspection. To address the limitations of traditional methods and enhance the effectiveness of automated detection, the article proposes an improved deep learning-based detection algorithm, named AMP-YOLO, built upon the lightweight and efficient YOLOv11n architecture. The proposed AMP-YOLO algorithm introduces a series of enhanced design to improve detection accuracy. Firstly, the paper replaces the downsampling of YOLOv11n with an adaptive downsampling module. Unlike conventional down-sampling methods of lose critical spatial details, ADown adaptively preserves contextual and edge structure information, particularly for small objects, by combining average pooling and max pooling strategies alongside lightweight convolutional blocks. This ensures that fine-grained features essential for identifying small parasites are retained during the feature extraction process. Secondly, it integrates a Mixed Local Channel Attention (MLCA) mechanism. MLCA enhances the network’s ability to capture and differentiate between local and global contextual features, allowing the model to refine overlapping features and improve differentiation of closely clustered or partially occluded parasites. As a result, the precision is effectively improved, addressing a common shortcoming in many object detection frameworks where subtle or obscured targets are easily overlooked. Finally, we replace the original CIoU loss function in YOLOv11n with the more robust MPDIoU loss function. MPDIoU introduces the distribution characteristics of center point deviation to avoid gradient vanishing, improve bounding box positioning accuracy, and enhance performance in scenes with occluded objects. This way contributes significantly to the model’s precision and recall, especially when detecting multiple parasites within densely populated regions. To comprehensively evaluate the performance of AMP-YOLO model, a series of extensive experiments were carried out using a self-built, custom-annotated dataset specifically curated for the task of parasitic detection in aquaculture. The dataset reflects real-world challenges common in practical settings, including varying parasite overlap and significant size differences that often confuse standard detection algorithms. The results demonstrate that AMP-YOLO significantly outperforms the baseline in all key aspects. Specifically, it achieves a mAP@0.5 4.8% higher than that of the original model (indicating greater overall detection accuracy), a 1.6% increase in precision (reducing false positives while correctly identifying true positives), and a 3.2% improvement in recall (capturing more true instances, particularly small, occluded, or partially visible parasites). These improvements confirm the high accuracy and completeness of the improved algorithm in parasite detection scenarios with size imbalance and target occlusion. The AMP-YOLO algorithm represents a better advancement in automated detection of fish parasites, offering a robust, efficient, and intelligent solution for aquaculture health monitoring. Through architectural innovations that improve feature extraction and localization accuracy, the model enables rapid, precise identification of fish parasitic in complex aquatic environments, even with severe overlap, occlusion, or visual ambiguity. This significantly reduces reliance on manual inspection, minimizing diagnostic errors caused by human subjectivity or limited experience. The enhanced accuracy of AMP-YOLO supports timely and targeted medical intervention, helping aquaculture practitioners quickly and accurately identify parasites, provide a basis for effective parasitic diseases prevention and control,, and ultimately promote the intelligent development of fish health management.