Hybrid BILSTM Network for Improving Crop Pest Classification
S. Pournima, C Priyatharsini, G. Kirubasri, J. Manikandan
Abstract
Insect infestations cause considerable monetary losses for the agricultural industries that produce rice, millet, maize, legumes, sugarcane, garbanzo, and onions. It is essential to determine the species of the bug as as quickly as possible in order to minimize the amount of money lost to it. Nevertheless, farmers have a difficult time discriminating between the numerous types of agricultural insects because they lack the experience and understanding to do so. In order to find a solution to this problem, one option is to make use of convolutional neural networks, often known as CNNs, together with other computer-based techniques. CNNs are useful for the identification of many images because of their capacity to automatically learn features that are dependent on the information from the data themselves. Within the scope of this study, we proposed a hybrid BiLSTM network as a potential solution to this problem. A pretrained model and a layer of bidirectional long short-term memories (BiLSTM) that takes temporal information into consideration are included in the architecture that has been recommended.