Crop pest classification based on deep convolutional neural network and transfer learning

The growth of most important field crops such as rice, wheat, maize, soybean, and sugarcane are affected due to attack of various pests and the crop production is reduced due to different types of insects. The classification and identification of all types of crop insects correctly is a difficult task for the farmers due to the similar appearance in the earlier stage of crop growth. To address this issue, Convolutional neural network (CNN) with deep architectures is being applied as it performs automatic feature extraction and learns complex high-level features in image classification applications. This study proposed an efficient deep CNN model to classify insect species on three publicly available insect datasets. The National Bureau of Agricultural Insect Resources (NBAIR) dataset used as first insect dataset that consists of 40 classes of field crop insect images, while the second and third dataset (Xiel, Xie2) contains 24 and 40 classes of insects respectively. The proposed model was evaluated and compared with pre-trained deep learning architectures such as AlexNet, ResNet, GoogLeNet and VGGNet for insect classification. Transfer learning was applied to fine-tune the pre-trained models. The data augmentation techniques such as reflection, scaling, rotation, and translation are also applied to prevent the network from overfitting. The effectiveness of hyperparameters was analysed in the proposed model to improve accuracy. The highest classification accuracy of 96.75, 97.47, and 95.97% was achieved in proposed CNN model for NBAIR insect dataset (40 classes), Xiel (24 classes) insect dataset and Xie2 (40 classes) insect dataset respectively. The results demonstrated that the proposed CNN model is effective in classifying various types of insects in field crops than pre-trained models and can be implemented in the agriculture sector for crop protection.