Pest detection on Traps using Deep Convolutional Neural Networks

It is commonly known that toxic pests have a negative influence on the production process and ultimately on the product quality of many industries. Therefore, it is reasonable to consider pest detection a crucial task in these production procedures in order to make relevant pest management decisions. However, the challenge here is that localization and classification of different insect species are fairly difficult due to high similarity in features between them, and it is even more challenging when particularly dealing with those already caught on traps. Inspired by the achievement of the Deep Convolutional Neural Network (CNN), this paper proposes a method of identifying various types of trapped insect species by making prediction based on available images. Using a database of 200 pictures (from a confectionery factory) including approximately 3,000 insects of 6 kinds, the accuracy rates of detection and classification are about 84% and 86% respectively.