DESIGN OF A DEEP LEARNING BASED CROP DISEASE DETECTION IN SMALL HOLDER FARMING SYSTEMS

ORUCHE, TOCHUKWU WISDOM (2026) DESIGN OF A DEEP LEARNING BASED CROP DISEASE DETECTION IN SMALL HOLDER FARMING SYSTEMS. Other thesis, Godfrey Okoye University, Enugu.

Text TOOCHUKWU COMPLETE WORK.pdf Download (2MB)

Abstract

Crop diseases continue to pose a serious threat to agricultural productivity because timely diagnosis remains difficult when farmers rely solely on visual observation. The dependence on manual inspection often results in delayed identification and inaccurate disease recognition, which can reduce crop yield and increase economic losses. Although artificial intelligence has increasingly been explored for agricultural applications, there remains a need for practical systems capable of delivering reliable disease diagnosis under real-world conditions. This study introduces a multistage intelligent framework that automatically predicts and classifies crop diseases from digital images. The proposed framework first filters non-plant images using a plant verification model before identifying the crop species and subsequently detecting diseases specific to that crop. Model training was performed using the Crop Pest and Disease Detection (CCMT) dataset together with an additional manually compiled collection of non-plant images. Prior to training, image enhancement and augmentation procedures were implemented to improve model robustness and minimise the effects of dataset imbalance. EfficientNet-B0 was adopted as the feature-learning architecture because of its favourable balance between computational efficiency and predictive performance. Experimental evaluation demonstrated that the plant verification model achieved a classification accuracy of 96%, while the crop identification model produced an overall accuracy of 84.85%. The cassava disease classifier delivered the strongest predictive performance, whereas the maize classifier also achieved satisfactory results despite the visual similarity among several disease categories. Although the tomato disease model produced consistently high recall values, it was intentionally excluded from the deployed system following the established deployment criteria.

Item Type:	Thesis (Other)
Subjects:	Q Science > Q Science (General)
Divisions:	Faculty of Computing And Information Technology (FACIT)
Depositing User:	Uchenna Eneogwe
Date Deposited:	24 Jul 2026 14:59
Last Modified:	24 Jul 2026 14:59
URI:	http://eprints.gouni.edu.ng/id/eprint/6015

Actions (login required)

View Item