Article: Pest classification using hybrid deep learning with optimal feature selection technique Journal: International Journal of Ad Hoc and Ubiquitous Computing (IJAHUC) 2026 Vol.51 No.4 pp.217 - 234 Abstract: This research proposes an integrated methodology for pest classification in crops using a combination of internet of things (IoT) sensors, image processing techniques, and machine learning models. The process begins with the acquisition of crop images in the field through IoT sensors, which are stored in a database for subsequent analysis. The proposed method utilises publicly available datasets for image processing, employing a median filter to pre-process the captured images and extract essential attributes through Haralick features and histogram of gradient (HoG)-based feature extraction techniques. To optimise computational efficiency, the extended aquila optimisation (EaqO) algorithm is employed for selecting the most relevant features. The final step involves pest classification using an attention-based DenseRecurrent model, combining DenseNet-121, bidirectional gated bidirectional recurrent unit (BiGRU), and attention mechanisms. The experimental findings provide enhanced accuracy, specificity, precision, recall, and F-score in pest categorisation, demonstrating the efficacy of the suggested technique. Inderscience Publishers - linking academia, business and industry through research

Title: Pest classification using hybrid deep learning with optimal feature selection technique

Authors: Sathya Mariappan; Mareeswari Varathunga Pandian; Aruna Ramalingam; Solairaj Anandappan

Addresses: CSE Department, Nadar Saraswathi College of Engineering and Technology, Theni, India ' CSE Department, AMC Engineering College, Bangalore, India ' ECE Department, AMC Engineering College, Bangalore, India ' CSE Department, Sethu Institute of Technology, Virudhunagar, India

Abstract: This research proposes an integrated methodology for pest classification in crops using a combination of internet of things (IoT) sensors, image processing techniques, and machine learning models. The process begins with the acquisition of crop images in the field through IoT sensors, which are stored in a database for subsequent analysis. The proposed method utilises publicly available datasets for image processing, employing a median filter to pre-process the captured images and extract essential attributes through Haralick features and histogram of gradient (HoG)-based feature extraction techniques. To optimise computational efficiency, the extended aquila optimisation (EaqO) algorithm is employed for selecting the most relevant features. The final step involves pest classification using an attention-based DenseRecurrent model, combining DenseNet-121, bidirectional gated bidirectional recurrent unit (BiGRU), and attention mechanisms. The experimental findings provide enhanced accuracy, specificity, precision, recall, and F-score in pest categorisation, demonstrating the efficacy of the suggested technique.

Keywords: IoT-based image acquisition; pest classification; crop disease; attention-based deep learning; optimal feature selection; aquila optimisation.

DOI: 10.1504/IJAHUC.2026.153345

International Journal of Ad Hoc and Ubiquitous Computing, 2026 Vol.51 No.4, pp.217 - 234

Received: 12 Mar 2024
Accepted: 23 Mar 2025
Published online: 01 May 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Pest classification using hybrid deep learning with optimal feature selection technique

Keep up-to-date