Optimizing Wheat Ear Head Detection and Yield Estimation Through UAV Based Imagery

Accurate identification and counting of wheat heads are essential for reliable crop yield estimation. This study introduces a novel model, customized YOLOv8, specifically designed for wheat head detection in challenging agricultural environments using orthomosaic UAV imagery. Our end-to-end pipeline starts with orthomosaic images, identifying and extracting plots using YOLOv8, followed by plot tiling and detection on these tiles. The results are seamlessly stitched to reconstruct the original image with detected wheat heads. To enhance image quality-improving brightness, contrast, exposure, and reducing blur-Real ESR GAN was employed. Coupled with the Random Cut Out Method for occlusion handling, this significantly improved input image quality, facilitating accurate detection under challenging field conditions. Evaluation of the Mahycho dataset demonstrates promising results, with the model achieving a mean average precision (mAP) of 99.2% on plot detection and 83.4% accuracy on wheat head detection, surpassing existing benchmarks. The model exhibits robust performance under adverse conditions, maintaining high R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> values, and retains accuracy across different growth stages in field-collected datasets. Overall, the customized YOLOv8 model offers precise wheat head detection, advancing crop monitoring and management in real-world agricultural applications.