Estimation of infestation level of diamondback moth and cabbage cluster caterpillar based on images

Abstract

Diamondback moth, Plutella xylostella (L.) (Lepidoptera: Plutellidae), and cabbage cluster caterpillar, Crocidolomia pavonana (F.) (Lepidoptera: Pyralidae), are the two most important insect pests of cabbage in Indonesia. Research was conducted with the objectives to assess level of infestsation of the pests based on digital images. Infestation levels were classified into five categories: healthy/negligible, low, moderate, severe and very severe. The data used in this research were images of cabbage that were healthy and attacked by Plutella xylostella and/or Crocidolomia pavonana. Images of cabbages with crop and without crop were 476 and 24 images respectively. The images were divided in two sets: training set and test set with percentage are 80% and 20% respectively. Ten-fold cross-validation was used to find the best model for classifier. Estimation of infestation level initially was made on individual plants. Cabbages with crop and without crop were classified with probabilistic neural network (PNN) based on four Haralick features, namely: contrast, correlation, dissimilarity, and homogeneity. Gray level co-occurrence matrix (GLCM) was used to extract texture features. The research showed that the average accuracy from PNN as a classifier for classifying cabbage with crop and without crop was 92.4%. Cabbage with crop was segmented, and after that the crop area was detected with randomized hough transform (RHT). The detected crop then were grouped into three categories: good, good enough and not good. Our research showed that estimation of crop area with RHT had accuracy of 89.13 %. Image enhancement was then applied to the crop area through shadow removal and smoothing. The level of damage was estimated based one proportion of holes on the crop area by applying erosion operations and thresholding with Otsu. Estimation of level of damage for cabbage with crop had accuracy of 76.1%. Estimation of level of damage for cabbage with crop and without crop had accuracy of 75%. Having obtained the level of damage for each plant, the final stage of this research was to assess level of caterpillar infestation in cabbage fields. Error arising from this estimation was 3.25%. Results of this research could be used by pest observers or extension agents in making pest management decisions.