Effectiveness of convolutional neural network models in classifying agricultural threats

Abstract

Smart farming has recently been gaining traction for more productive and effective farming. However, pests like monkeys and birds are always a potential threat for agricultural goods, primarily due to their nature of destroying and feeding on the crops. Traditional ways of deterring these threats are no longer useful. The use of highly effective deep learning models can pave a new way for the growth of smart farming. This study aims to investigate the manner in which deep learning convolutional neural network (CNN) models can be applied to classify birds and monkeys in agricultural environments. The performance of CNN models in this case is also investigated. In this regard, four CNN variants, namely, VGG16, VGG19, InceptionV3 and ResNet50, have been used. Experiments were conducted on two datasets. The experimental results demonstrate that all the models have the capability to perform classification in different situations. Data quality, parameters of the models, used hardware during experiments also influence the performance of the considered models. It was also found that the convolutional layers of the models play a vital role on classification performance. The experimental results achieved will assist smart farming in opening new possibilities that may help a country’s agriculture industry, where efficient classification and detection of threats are of potential importance.