Weed detection using image processing under different illumination for site-specific areas spraying

• We adopt component in the color model space gray images.
• We combine two methods which identify the center line of the crop rows quickly.
• We improve the Weeds Infestation Rate (WIR) to reduce computational complexity.
• We provide online variable spraying decision.

Large area bold type spraying of chemical herbicide is not only a waste of herbicides and labor, but also leads to environmental pollution and food quality problems. Traditional methods have the problems of high light and sample quality etc requirements. Therefore, accurately identifying weeds and precisely spraying are important strategies for promoting agricultural sustainable development. To avoid the influence of different illumination on images, this paper adopts the color model and then proposes component to gray images; the vertical projection method and the linear scanning method are combined to quickly identify the center line of the crop rows; the classic Weeds Infestation Rate (WIR) is modified to decrease the computational complexity and the improved horizontal scanning method is taken to calculate within cells; finally, Modified Weeds Infestation Rate (MWIR) is used to realize real-time decision through the minimum error ratio of Bayesian decision under normal distribution. The experimental results show that the accuracy of this algorithm is 92.5%, which exceeds the BP algorithm and SVM algorithm.