Classification of Insect Pest Using Transfer Learning Mechanism

Classification of Insects is one of the most vital and essential research, which needs to be done for, various factors like, for the protection of crops in the agricultural sector. The identification of crop pests is a difficult problem since, pest infestations cause significant crop damage and quality degradation. The majority of insect species are quite similar to one another that makes the task of detection of the insect on field crops like rice, soybeans, and other crops more challenging than a normal detection of objects. Currently, classifying insects manually is the major method used to distinguish them in crop fields, but this is a time - consuming and expensive operation. Considering the advancements in the field of deep learning, we propose to use a pre-trained network model trained on a millions of images of ImageNet dataset to do the classification task using transfer learning mechanism. An extensive experimentation was done using various pretrained models like VGG, inception, xception, ResNet, MobileNet, DenseNet and efficient net. Various insect datasets were used for classification task and model was fined tuned using transfer learning. The EfficientNet B7 model has achieved the highest accuracy 70%, 98% and 99% on IP102 (102 classes), Xie (40 classes) and Kaggle village Synthetic dataset (10 classes) respectively.