Disease Detection in Solanaceous Crops using One-Stage Detectors

Asar Khan; Syafeeza Ahmad Radzi; Norihan  Abdul Hamid; Azureen Naja  Amsan; Norazlina  Abd Razak; Shahid  Rahaman

doi:10.37934/ard.134.1.113

Authors

Asar Khan Machine Learning and Signal Processing (MLSP) Research Group, Fakulti Teknologi dan Kejuruteraan Elektronik dan Komputer (FTKEK), Universiti Teknikal Malaysia Melaka, 76100, Durian Tunggal, Melaka, Malaysia
Syafeeza Ahmad Radzi Machine Learning and Signal Processing (MLSP) Research Group, Fakulti Teknologi dan Kejuruteraan Elektronik dan Komputer (FTKEK), Universiti Teknikal Malaysia Melaka, 76100, Durian Tunggal, Melaka
Norihan Abdul Hamid Machine Learning and Signal Processing (MLSP) Research Group, Fakulti Teknologi dan Kejuruteraan Elektronik dan Komputer (FTKEK), Universiti Teknikal Malaysia Melaka, 76100, Durian Tunggal, Melaka, Malaysia
Azureen Naja Amsan Machine Learning and Signal Processing (MLSP) Research Group, Fakulti Teknologi dan Kejuruteraan Elektronik dan Komputer (FTKEK), Universiti Teknikal Malaysia Melaka, 76100, Durian Tunggal, Melaka, Malaysia
Norazlina Abd Razak Machine Learning and Signal Processing (MLSP) Research Group, Fakulti Teknologi dan Kejuruteraan Elektronik dan Komputer (FTKEK), Universiti Teknikal Malaysia Melaka, 76100, Durian Tunggal, Melaka, Malaysia
Shahid Rahaman Department of Computer Science, University of Buner, Pakistan

DOI:

https://doi.org/10.37934/ard.134.1.113

Keywords:

Agriculture, crop disease, deep learning, object detection, one-stage detector

Abstract

Agriculture plays a crucial role in sustaining and ensuring the continuous food supply. Crop disease may cause the negative impact on agriculture due to the decrease of yield production. Machine vision technology such as object detection can overcome the issue of early disease detection with more efficient way compared to the conventional method such as manual observation. This study utilizes two one-stage detectors namely YOLOv8 and SSDLite-MobilenetV3 to analyze the efficiency and accuracy of both models to perform crops disease detection. A total of 23 species of plants dataset are used taken from PlantVillage dataset. The datasets are divided into 70:20:10 ratio which results in total of 9,936 for training, 1,414 for validation, and 510 for testing images are used. The result shows that YOLOv8 has better performance with 86% accuracy compared to 82% for SSDLite-MobilenetV3. YOLOv8 also surpassed SSDLite-MobilenetV3 in terms of inference time by 76.6% faster with 8.2ms and 35.5ms respectively.