Intra-row weed density evaluation in rice field using tactile method

Abstract

• The first attempt to evaluate the density of intra-row weeds in a paddy field based on tactile perception. • A variety of tactile expressions of weed density were obtained. • A novel feature extraction scheme based on the Hilbert-Huang Transform was proposed. • The weed density evaluation model was established based on the feature scheme. Accurate evaluation of weed density is crucial for effective utilization of herbicides, improvement of rice quality, and reduction of herbicide dosages. The application of visual methods is disadvantageous because intra-row weeds are blocked by the canopies of adjacent rice plants. Therefore, an innovative tactile sensing method is proposed. A flexible gasbag filled with special microstructures distributed over its surface was developed. The tactile data of weed density were generated through contact between the microstructures and weeds, and the data were measured using the voltage value of a barometric sensor mounted inside the gasbag. The tactile time series was processed using fractal theory and Hilbert–Huang transform (HHT), and the discriminating features of the weed density were acquired. The discriminating features were input into a neural network to train a weed density classifier to evaluate the weed density. The results of the feasibility experiment demonstrated that the evaluation accuracies for high-density, medium-density, and low-density weeds were 95.4%, 91.8%, and 87.9%, respectively, with an average accuracy of 91.7%. The field validation test demonstrated that the visual-based method had an average classification accuracy of 64.17%, whereas the proposed method had an average accuracy of 77.04%, experimentally demonstrating superior accuracy over the image-based method.