Aquaculture plays an important role in ensuring global food security, supporting economic growth, and protecting natural resources. However, traditional methods of monitoring aquatic environments are time-consuming and labor-intensive. To address this, there is growing interest in using computer vision for more efficient aqua monitoring. Fish detection is a key challenging step in these vision-based systems, as it faces challenges such as changing light conditions, varying water clarity, different types of vegetation, and dynamic backgrounds. To overcome these challenges, we introduce a new model called AquaYOLO, an optimized model specifically designed for aquaculture applications. The backbone of AquaYOLO employs CSP layers and enhanced convolutional operations to extract hierarchical features. The head enhances feature representation through upsampling, concatenation, and multi-scale fusion. The detection head uses a precise 40 × 40 scale for box regression and dropping the final C2f layer to ensure accurate localization. To test the AquaYOLO model, we utilize DePondFi dataset (Detection of Pond Fish) collected from aquaponds in South India. DePondFi dataset contains around 50k bounding box annotations across 8150 images. Proposed AquaYOLO model performs well, achieving a precision, recall and mAP@50 of 0.889, 0.848, and 0.909 respectively. Our model ensures efficient and affordable fish detection for small-scale aquaculture.