Deep Learning-Based Crop Pest Detection and Classification using DenseNet Technique

Abstract

A correct classification of species will help in choosing a proper strategy for crop management, but designing self-operating solution is also problematic due to high similarity among species. The agriculture field has immense potential for improvement of needs of food and provides healthy food.  Yield loss is a major problem in the agriculture field due to attack of various insect pest. Farmers often face challenges in recognition of crop insects as a significant portion of crop is damaged. Early information of pest attack can help farmers to reduce damage and enhance productivity of crops. The quality of crops degraded due to pest attack. Nowadays, researchers execute a deep learning approach to classify various kinds of insect pests by usi practically. This paper aims to detect insect pests in crops using a deep learning technique. DenseNet model is used in respect to recognize insect pests. The model neck creates feature pyramids for feature extraction to obtain a single stage image randomly sized for input, output with proportionate-size feature map at various levels. IP 102 dataset is utilized for the purpose of training, and testing of model. The total images of 292 rice leaf caterpillar and 669 for rice leaf roller class were used for training and testing of testing respectively. Experiments were conducted on two classes namely rice leaf caterpillar and rice leaf roller. The highest classification accuracy of 87.90% achieved. The comparison also presented comprising accuracy of different other models with proposed DenseNet model. The classification results of model are utilized to recognize insect pests in crops.