Classification of Winter Wheat and Weeds by RGB Images and Deep Learning Approach

Precision weed management is essential in modern agriculture, aiming to effectively control weeds while minimizing crop damage. In this study, we developed a deep learning-based approach using color images to differentiate winter wheat from four common broadleaf weed species: annual bugloss, bastard cabbage, hoary cress, and flixweed—resulting in five classification categories. A dataset of ۵۴۲ field images was collected under natural conditions, resized to ۲۹۹×۲۹۹ pixels, and normalized to a [۰,۱] range to standardize inputs. The Xception network, pre-trained on ImageNet, was employed as the base model. Using transfer learning, the network was fine-tuned and improved with additional layers to optimize performance for this specific task. The improved Xception model achieved a classification accuracy of ۹۷% on the test set, with a loss of ۳.۲۵ and F۱-scores between ۹۳% and ۱۰۰%. These results establish a robust framework for precision weed management, enabling reduced herbicide use, lowering environmental impact, and promoting improved wheat yield and quality.

کلیدواژه ها:

Winter wheat ، Broadleaf weed ، Deep learning ، Classification ، RGB images