Cotton Crop Leaf Disease Detection System Using Machine Learning Approaches to Improve Efficiency”

Abstract

India's economy heavily depends on agriculture. Over 70% of people in India make their living from agriculture. Accurate and prompt diagnosis of illnesses that harm crops is one of the biggest challenges facing the agriculture sector. Diseases affect crop quality and have the power to destroy whole hectares of agricultural output, costing farmers a lot of money.

Current diagnosis methods need the presence of highly experienced professionals and take a lot of time to examine the damaged crop, understand the symptoms, identify the illness, and provide effective remedies. Due to the limitations of these methodologies, researchers are now looking for other approaches to early illness detection and classification. Addressing food security may be facilitated by smart farming and adequate infrastructure.

In recent years, machine learning has demonstrated tremendous potential in identifying and categorising trends in linked academic disciplines. The goal of the current study is to evaluate the accuracy, precision, recall, and training time of traditional machine learning techniques like the Support Vector Machine (SVM) and random forest against the performance of convolutional neural network (CNN) methods and architectures like Inceptionv3, VGG16, and RasNet50 with data augmentation and transfer learning. The models were trained with the use of a manually gathered database from a farm and a government organisation, which had four distinct classes of photos, including healthy plants. The highest performing model was the Inceptionv3 architecture of CNN with transfer learning, which achieved an overall accuracy of 94 percent and met the demand for a more reliable and effective classification model. Additionally, when the quantity of training data rose, it was shown that the performance of the developed models increased.

The outcomes obtained using transfer learning algorithms on CNN architectures are extremely encouraging, and they may be further refined to create a thorough leaf disease diagnosis system that can function in a real-world environment. As a result, it may enable the agricultural community to recognise problems and start prompt treatment without the intervention of qualified specialists.