Estimation of Whitefly Population Density on Vegetables Using Watershed Segmentation

Abstract

Witheflies Bemisia tabaci and Trialeurodes vaporariorum are two important pests on vegetables. This research proposed a new method of estimating population density of whiteflies using watershed algorithm. Estimate of population density was obtained by counting the number of pests found on leaves. Watershed algorithm is highly appropriate for segmentation of whitefly, because this algorithm works properly on objects which has low gray level variation and overlapping areas. The research methodology consisted of image acquisition, image preprocess, image enhancement segmentation, calculation of whitely, calculation of infested leaf area, calculation of whitefly population density, and evaluation of infestation level estimates. Data used in this research were digital images of vegetable leaves infested by the whiteflies. Image pre-processing was used to reduce errors that might occur in the process of segmentation with watershed algorithm. Image enhancement with was aimed to reduce noises. Whitefly calculation was done after image segmentation. Infestation area was calculated by sampling leaf areas. The goal was to separate the non object areas so not to be included in the calculation process. The calculation results used to determine the level of pest population density. Whitefly densities were calculated using two methods: sampling area and leaf segmentation. The sampling area method was used when the observed images had a background difficult to separate from leaf objects, and the leaves were not in perfect condition such as folded, truncated, or blured. Whereas, the leaf segmentation method were used when the leaves images were in perfect condition, not bent or not blured, and the background was easy to separate. The process of testing was done using images grouped into three levels: low, moderate, and severe. Results of both methods then were with the estimates by expert. Our research showed that estimation of whitefly population density had an error of 12,5%. The sampling area method was chosen to estimate whitefly density because of its closeness to the expert estimates. Based on this method, the pest population infetations were grouped into three levels: low (≤ 10%), moderate (> 10% until ≤ 30%), and severe (> 30%). The system of estimating whitefly density developed through this research is expected to help pest observers in executing pest population monitoring.