Identification and classification model of wheat and weed based on improved faster R-CNN

Weed is one of the major causes for wheat yields reduction. Accurate and efficient classification and identification of wheat and weed could assist farmers to control weed effectively, which is of great significance for food security. This study aims to develop a novel model utilizing transfer learning methods and Faster R-CNN algorithm, combined with advanced image processing techniques to achieve automated detection, classification and identification of wheat and weed. First, dataset containing multiple species is constructed by collecting a large number of wheat and weed sample images. Then an improved deep learning model, namely the two-stage object detector Faster R-CNN, is trained and optimized to improve classification accuracy and generalization capability. Meanwhile, two traditional models different from Faster R-CNN are trained on the same dataset and evaluated using the same MS COCO metrics for comparison. Experimental results demonstrate that the proposed system exhibits higher accuracy compared to traditional detection models, and possesses outstanding advantages in more complex recognition environments. Thereby the system could provide significant practical value for improving wheat production quality in real-world scenarios.