Object Detection of Alternanthera Philoxeroides at Seedling Stage in Paddy Field Based on Faster R-CNN

Abstract

Weed competing with the rice seedlings for resources, which also provided a better condition for pests and diseases. Targeted spraying herbicide of weed was the key to greatly reduce pesticides and reduce pollution. Weed category information and position detection were the basis of the field intelligence management. In order to realize the automatic recognition and position detection of weed from the complicated background and the natural light in the paddy field, this paper presented a new recognition method for weeds at seedling stage in paddy fields using Faster R-CNN (convolutional neural network). In this proposed method, transfer learning and target detection framework with Faster R-CNN were combined, and transfer the feature extractor of pre-training network model VGG16 and AlexNet. The result indicated that the recognition accuracy of Faster R-CNN with backbone network of VGG16 is better than the that of AlexNet, which could reach 82.47%.