Classification of Weeds and Crops using Transfer Learning

Abstract

Weed control is critical in agriculture to ensure productivity and efficiency. Chemical herbicides are the most commonly used strategy for weed management. However, excessive pesticide use has wreaked havoc on the ecosystem while also posing a health risk. These worries have triggered a rise of research interest in finding new weed-control strategies. A computerised machine learning based system that can distinguish weeds and crops in digital photos could be a cost-effective approach to reduce pesticide use. This study proposes a method for automatically classifying images into crops and weeds based on Transfer learning. This study employs different transfer learning algorithms to classify the images into multiple classes. The dataset contains 347 images representing the crop tomato and two weed species. To increase the variability in the dataset, data augmentation was used. In comparison to the other models, MobilenetV2 produced the best results. A 10-fold cross validation technique was used to validate the results of MobilenetV2.