Deep-Learning Model for Wheat Disease and Pest Classification

---

Abstract

Severe pests and diseases in wheat are caused by global change and natural disturbances. This results in a significant loss of yield and quality. Therefore, ecological problems will be solved by early detection of dis-eases and pests. In prevailing techniques, different types of wheat diseases and pests were not studied com-pletely. Hence, a Deep-Learning (DL) framework for wheat field disease and pest classification in leaves via satellite images using Non-monotonic Correlated-Extreme Learning Machine (NC-ELM) is proposed. Primarily, the farm field’s satellite images are given as input to the proposed framework. Afterward, to re-move unwanted noises, the images are pre-processed using Wavelet Spearmann Rank Transform (WSRT). Next, by utilizing the Adaptive Irregular SegNet (AIS) algorithm, the leaf regions are segmented. Then, to separate patches, the slicing of images is evaluated. After that, the graph is constructed via the Cosine Simi-larity induced Kruskal’s Minimum Spanning Tree (CSK-MST) algorithm. Here, the pests are indicated in highlighted points; then, the features are extracted. Moreover, the features are extracted from the other re-gions. The optimal features are selected from the extracted features using the Parallel Group Gazelle Opti-mization Algorithm (PG-GOA). Lastly, the resultant optimal features are fed into the NC-ELM classifier. Hence, NC-ELM classifies pests and diseases. Experiments were conducted for the proposed technique with the conventional approaches, which proved the proposed one’s efficacy.

---

Keywords

Wavelet Spearmann Rank Transform (WSRT), Adaptive Irregular SegNet (AIS), Cosine Similarity induced Krus-kal’s minimum spanning tree (CSK-MST) algorithm, Parallel Group Ga-zelle Optimization Algorithm (PG-GOA), and Non-monotonic Correlated-Extreme Learning Machine (NC-ELM).