Abstract:
"This thesis discusses the use of machine learning and image processing to classify soil types in Sri Lanka.
The goal is to promote sustainable agricultural practices through the use of technology. With the rapid
advancement of digital tools, there is significant potential to improve agricultural productivity by
incorporating intelligent systems into daily farming activities. This research involves creating a mobile
application that utilizes deep learning algorithms to analyze soil images taken by smartphones, thus
identifying and categorizing different soil types.
In collaboration with the Peradeniya Agriculture Department, a custom dataset representing Sri Lankan
soils was developed. It includes multiple classes such as red, reddish-brown, black, and clay soils. Each
class consists of 1300 images, meticulously annotated and processed to train a Convolutional Neural
Network (CNN). The model showed an accuracy range of 87% to 90%, demonstrating its effectiveness in
real-world conditions.
The application resulting from this research not only recommends suitable crops based on the identified
soil type but also provides guidance on optimal planting strategies. This enables precision farming,
allowing farmers to maximize yield while conserving resources. Additionally, the project extends the
potential of this technology to diagnose soil-related crop diseases, laying the foundation for a
comprehensive agricultural advisory tool.
This study contributes to the existing knowledge by showing the feasibility of using machine learning for
soil classification in Sri Lanka, offering a scalable model that can be adapted to other agricultural
environments. The research emphasizes the transformative potential of integrating modern technology
with traditional farming practices, with the aim of attracting young people to agriculture and supporting
sustainable development goals."
