Abstract

It is well recognized that one of the main factors causing harm to agricultural crops of economic importance is insect pests. The availability of food and a stable agricultural sector depends on accurate agricultural pest prediction, which is largely dependent on the classification of insect pests. Insect pest recognition is time-consuming and expensive since it mostly depends on the special expertise of agricultural specialists due to a wide variety of species of pests and tiny variations between species. This research proposes a robust FS (Feature Selection) and HDL (Hybrid Deep Learning) model for the efficient recognition of pests using IP102 dataset in CC (Cloud Computing) structure. Initially, the pre-processing is done by Z-score normalization for removing the noises for improving the classifier’s execution. Second, the FS is applied by the technique depends on MCSA (Modified Cuckoo Search Algorithm). And finally, the Hybrid DL model is proposed for the efficient recognition of pests. Here the Granular Neural Network (GNN) is hybridized with the Faster R-CNN (Region-Convolutional NN) for the improved execution of the classifier. Thus the result demonstrates that the proposed recognition technique is implemented with the IP102 dataset and which is evaluated with the metrics as accuracy, precision, recall and F-measure. As a result of analysis, it is identified that the proposed technique has high execution results than the existing recognition models.

Authors

P. Bharathi, K. Dhanalakshmi
Kongunadu Arts and Science College, India

Keywords

Agricultural Pests, Pre-Processing, Feature Selection, Z-score Normalization, Modified Cuckoo Search Algorithm (MCSA), HDL (Hybrid Deep Learning) Model, Granular Neural Network (GNN)