Weed Detection and Treatment by RGB System - Case of the FPL Smart Farm

Abstract

Strawberry is one of the most important crops in the Loukkos area in Morocco; it exceeds 80% of the national production. This crop is very sensitive and requires well-controlled climatic conditions, specific inputs and meticulous monitoring to avoid any uncertainty in yield. Our work consists in treating the side related to the weeds, which can cause considerable losses of yield and disturb plant development. The danger of these weeds in this culture is the strong competition with the needs of the culture (water, nutrients, light...). The objective of our work is to make a state of the art by bibliographic systemic analysis and propose intelligent methods for detection and treatment of weeds based on RGB images. Most of the works are based on image acquisition mainly by smartphones or unmanned aerial vehicles (UAVs), and some by professional cameras. The image processing can be done by deep learning: through convolutional neural networks (CNN), support vector machine (SVM), or local binary pattern (LBP). Other methods mentionned allow detecting and controlling weeds using a robot that determines the undesirable plants by image processing (via deep learning), records their location and proceeds to a weeding or a localized treatment using herbicides. These methods allow decreasing labor dependency, a better treatment efficiency, and a lower herbicide use up to 90%. We thus propose a best suitable method to detect and treat weeds in an intelligent way in the "Smart Farm" of the Polydisciplinary Faculty in Larache. This method consists in using fluorescent paint at the stem base of the strawberry plant in order to differentiate the main crop from any undesirable plant. The robot that takes RGB images under UV rays considers as "weed" any plant that does not contain fluorescent paint.