A novel automated image analysis method for counting the population of whiteflies on leaves of crops

Abstract

Counting the population of insect pests is a key task for planning a successful integrated pest management program. Most image processing and machine vision techniques in the literature are very site-specific and cannot be easily re-usable because their performances are highly related to their ground truth data. In this article a new unsupervised image processing method is proposed which is general and easy to use for non-experts. In this method firstly a hypothesis framework is defined to distinguish pests from other particles in a captured image after texture, color and shape analyses. Then, the decision about each hypothesis is made by estimating a distribution function for sizes of particles which are presented in the image. Performance of the proposed method is evaluated on real captured images that belong to plants in green houses and farms with low and high densities of whiteflies. The obtained results show the greater ability of the proposed method in counting whiteflies on crop leaves compared to adaptive thresholding and K-means algorithms. Furthermore it is shown that better counting of the pest by proposed algorithm not only doesn't lead to extracting more false objects but also it decreases the rate of false detections compared to the results of the alternative algorithms.