Using video processing to classify potato plant and three types of weed using hybrid of artificial neural network and partincle swarm algorithm

Fighting weed is an effective way to increase crop yield, which can be achieved with mechanical methods or spraying herbicides. Although the latter type is the most used, it can cause environmental pollution and food poisoning. For this reason, precision farming technologies are applied to minimize the required amount of herbicides, spraying only in small areas where weeds appear. The aim of this study is to locate and identify potato plants and three common types of weeds (Chenopodium album, Secale cereale L, and Polygonum aviculare L.) using a novel machine vision system. This system comprises two main subsystems: (1) a video processing subsystem capable of detecting green plants in each frame; and (2) a machine learning subsystem to classify weeds and potato plants. A hybrid approach, consisting of artificial neural networks (ANN) and particle swarm optimization algorithm (PSO), is used in classification. This approach is able to optimize the number layers, neurons per layers, network functions, weights and bias. Image capture was performed in four farms of Agria potato variety located in the Iranian province of Kermanshah, under controlled lighting conditions using white LED lamps. After shooting, plants were segmented and 30 color, texture and shape features were extracted from each one. Then, a decision tree is used to select the 6 most significant features, in terms of difference between potato plants and weeds. Finally, ANN-PSO method is applied to classify the inputs as potato plants or weeds. A comparison is performed using a Bayesian classifier. The experimental results show that ANN-PSO and Bayesian achieved an accuracy of 99.0% and 71.7%, respectively, on the training set, and 98.1% and 73.3%, respectively, for the test set. These results indicate that a precise site-specific sprayer can be designed using the proposed approach, optimizing the use of herbicides in precision farming.