Efficient Net: Identification of Crop Insects Using Convolutional Neural Networks

In India agriculture offers employment to more than 50% (2018) of the workforce. Agricultural output is, however, affected by crop insects and the government has put several mechanisms in place to address this issue. Insect classification is a crucial part of this process to address insect infestation. Traditionally, insects have been identified by trained taxonomists who have wide and varied experience in detecting insects. However, satisfying the requirement of having qualified and competent taxonomists to identify insects based on their morphological characteristics is not a viable method for the geography of India and many parts of the world. Hence, to alleviate this situation, we have used Transfer Learning model to detect the insect. In this work, EfficientNet B0 model has been used to classify and recognise the insects. K-fold cross validation has been applied to make the models more robust and to make the prediction scores more dependable. The EfficientNet B0 model was able to surpass the benchmark accuracy and has improved classification and recognition of insects, which when used in the early stages of insect attack can be used to employ appropriate countermeasures effectively.