Machine learning-based mobile application for weed detection in paddy fields

Abstract

In the context of Sri Lanka, where agriculture, particularly paddy cultivation, plays a crucial role, farmers face significant challenges due to weed infestation. Unlike some other countries that have embraced machine learning technologies to address these issues, Sri Lanka has yet to adopt such advanced solutions. To tackle the pervasive weed problem, a research initiative was undertaken to develop a mobile application capable of identifying weed types. The methodology involved utilizing Convolutional Neural Network (CNN) pre-trained models, namely ResNet-50, Inception-v3, and VGG16, along with the Google Colab platform for training the dataset. Among the three models, VGG16 demonstrated the highest accuracy, making it the chosen model to further the research. The primary goal was to achieve a superior level of accuracy in detecting weed species in rice fields. The research team focused on delivering a mobile application with a high level of accuracy to identify and classify weeds in paddy fields. The integration of advanced technologies, such as IoT and machine learning, aimed to provide Sri Lankan farmers with an efficient and effective tool to combat weed-related challenges in their agricultural practices.