ABSTRACT Paddy is a food security crop in over 150 countries and one‐third of humanity's food depends on pest‐free paddy crops growing in the field and pests often increase the farmer's poverty level. One of the major goals of this research is to test the efficiency of artificial intelligence, such as deep and machine learning, for automatic classification of paddy pests. Among the wide varieties of pests found in the paddy fields of Bangladesh, six major paddy pests were selected for this research. Manual inspection and conventional machine learning methods, the two main elements in most existing pest identification systems, are unfeasible due to scalability and inaccurate practices under the various field environment they are operating. In addition, these drawbacks are tackled in this work by developing and analysing traditional models against deep‐learning concepts. It is evidenced that some models like VGG16, ResNet50, MobileNet, Inception‐v3 and a custom convolutional neural network (CNN) are deep learning models that are proving to be better than conventional machine learning models. With 99.46% accuracy, the custom CNN model outperformed the other models at the same algorithm. While k‐NN (k‐nearest mean), support vector machine, Decision Tree and Random Forest models resulted in 92.22%, 94.81%, 93.33% and 97.41% accuracy consequently. The deep learning models textbook pests based on visual features like shape, colour and texture and utilised both real‐time and augmented pictures. Deep learning served as a self‐assessment priority over conventional machine learning. Hence, the custom CNN model is proposed and its efficiency is measured in terms of accuracy, precision, recall and F1‐score.
Islam et al. (Thu,) studied this question.