Cotton Boll Detection through Deep Learning Techniques

The search for more efficient cotton harvesting methods has intensified due to the labor-intensive nature and high losses associated with current machine harvesting practices. In response, the exploration of cotton harvesting robots has emerged as a promising alternative. However, a significant challenge arises in accurately recognizing cotton bolls amidst interference from the sky. To tackle this challenge, the study turns to the capabilities of Convolutional Neural Networks (CNNs), specifically focusing on the CNN U-Net architecture. Impressively, the CNN U-Net model achieves a remarkable 99% accuracy in effectively segmenting cotton bolls from the sky. The study aims to assess the effectiveness of CNN U-Net in the intricate task of segmenting cotton bolls during robotic harvesting, emphasizing the importance of key metrics such as IoU, F1score, precision, and recall to evaluate accuracy, reliability, and segmentation speed. Through a comprehensive approach, the study harnesses the power of CNN U-Net for image segmentation, training and evaluating the model with critical metrics. The obtained results highlight the model’s exceptional performance, maintaining a consistent 99% accuracy in segmenting cotton bolls. These promising outcomes open avenues for the seamless integration of the CNN U-Net model into cotton harvesting robots, offering the potential for heightened efficiency and a reduction in errors during the harvesting process.