Developing Intelligent Weed Computer Vision System for Low-land Rice Precision Farming

Abstract

Weed infestation is one of the major problems facing rice production in Africa. Losses of rice caused by weeds yearly have been estimated at 2.2 million tons in Sub-Saharan Africa, the losses which are estimated at $1.45 billion. Weeds reduce the economic value of rice by causing an increase in the cost of production. Concerns have been raised on the health implication of herbicides, weeds seed in food crop and their effect on the environment, therefore, leading to the need for site-specific means of herbicide application to target only the weeds and ensure minimal seed contamination. This paper addresses these problems by the use Faster Regions with Convolution Neural Network (faster R-CNN) and Fuzzy Logic Controller (FLC) to develop an intelligent weed recognition system for better yield and return of investment in rice production in Sub-Saharan Africa. Faster R-CNN is a type of Artificial Neural Network (ANN) which uses convolutional features to map obtained features from an input image in order to identify the region of interest from the bounding box drawn around the weed image. As of the time of this research, the faster R-CNN method provides a faster means for real-time recognition as compared to other methods of ANN. The result of the recognition will be fed into the FLC to control the volume and time of spraying of the herbicides in low-land rice precision farming. The successful development and pilot testing of the anticipated intelligent computer vision system for rice weed control is expected to provide a faster and more efficient means of weed management for low-land precision farming for better food security in Sub-Saharan Africa.