PestDetector: A deep convolutional neural network to detect jute pests

Abstract

Widely known as the "Golden Fiber", jute is regarded as one of the most important and profitable crops in many countries including Bangladesh. Jute and jute-based commodities can bring a lot of foreign income and eventually boosts the overall economy of the country. However, many a time, jute production gets hindered due to many harmful pests and insects. Even though farmers identify and take actions against these pests following a manual procedure, it is often tedious and time-consuming. That is why it may be very beneficial to have a machine learning-based approach towards pest detection. This paper proposes a deep CNN model named "PestDetector"that can correctly identify 4 major types of jute pests (Field Cricket, Jute Stem Weevil, Spilosoma Obliqua, and Yellow Mite) with substantial accuracy. The work is done on a total of 2200 images separated into 3 categories: Training, Validation, and Testing. The model ultimately demonstrates 99.18% training accuracy and 99.00% validation accuracy. Additionally, the model's overall performance has been assessed using precision, recall, F1-score, and confusion matrix.