Computer vision, though capable of recognizing individual objects in images, requires high technical skills, computational resources, and may suffer from overfitting, limiting its applicability. Biodiversity monitoring employs generic models for habitat identification and population estimation, but YOLOv5 lacks data recovery capabilities. We propose an enhanced YOLO with a 2121-sized receptive field and a context improvement module to improve detection of faint targets. Fractional student psychology-based optimization is used for hyper-parameter tuning, demonstrating the efficacy of combining UAV images with deep learning for bird recognition, including bird decoys for more accurate data.