Welcome to the Bahria University DSpace digital repository. DSpace is a digital service that collects, preserves, and distributes digital material. Repositories are important tools for preserving an organization's legacy; they facilitate digital preservation and scholarly communication.
dc.contributor.author | Salman Ahmad Khan, 01-249212-013 | |
dc.date.accessioned | 2023-12-18T11:29:55Z | |
dc.date.available | 2023-12-18T11:29:55Z | |
dc.date.issued | 2023 | |
dc.identifier.uri | http://hdl.handle.net/123456789/16837 | |
dc.description | Supervised by Dr. Sumaira Kausar | en_US |
dc.description.abstract | Plant diseases pose a serious danger to crop production globally, leading to huge economic losses. According to forecasting, the world’s population will reach 9 billion by 2050, demanding seventy percent increase in food production. Presently, infectious diseases are responsible for an average crop potential reduction of approximately 40%, with numerous farmers in developing areas enduring yield losses of up to 100%. To address this challenge, smart agriculture is advancing by using IoT and AI methods for the identification and control the diseases on time to reduce yield loss. Several of these approaches rely on machine learning techniques that use visual data to promptly identify and diagnose diseases in real-time. As deep learning techniques continue to advance, innovative and novel models have emerged that harnessed the power of CNN for the multi-class classification of specific plant diseases. However, an emerging trend in vision-based deep learning, the using of ViTs, remains relatively unexplored in the realm of plant pathology applications. In this research, we proposed a groundbreaking model termed ”MobileViT,” which combines the strengths of traditional CNNs with ViTs to efficiently identify a wide array of plant diseases across various crops. The MobileViT model, as proposed, features a lightweight architecture, with just 1.6 million, 2.8 million, and 5.6 million trainable parameters in its XSS, XS, and S versions, respectively. This characteristic makes it particularly well-suited for IoT-driven smart agriculture applications. We evaluate the performance of MobileViT on the open-source Plant Village dataset. Our results indicate that the MobileViT network surpassed state-of-the-art techniques when utilized with the Plant Village dataset as a benchmark. Notably, the overall accuracy for classifying plant diseases exceeds 97.07%, 98.23%, and 98.47% for the three versions of MobileViT, respectively. This research marks a notable breakthrough in the realm of plant disease identification, providing a robust instrument to enhance crop health and bolster agricultural productivity during the era of smart agriculture. | en_US |
dc.language.iso | en | en_US |
dc.publisher | Computer Sciences | en_US |
dc.relation.ispartofseries | MS (DS);T-02072 | |
dc.subject | Light-Weight | en_US |
dc.subject | Approach | en_US |
dc.subject | Crop Disease | en_US |
dc.title | A Light-Weight Approach for Crop Disease Identification | en_US |
dc.type | Thesis | en_US |