Machine Learning Models in Precision Agriculture for Classification of Diseases

The bond between machine learning models and the agriculture field is becoming stronger with each passing day. The pre-trained AlexNet, ResNet, VggNet, and GoogleNet models commonly support various tasks in the precision agriculture field, such as disease detection, nitrogen status estimation, leaf area index identification, nutrient status measuring, finding pesticide requirements, measuring growth, etc. These tasks involve a huge amount of data, such as simple cell phone images, satellite imagery data, vegetation index data, images from UAVs, and sensor data from various soil and water sensors. Analyzing and finding insight from such data is a complex task. In this article, a framework is proposed using the transfer learning method for disease classification using manual CNN model along with the pre-trained AlexNet, ResNet, and GoogleNet. The 5450-image dataset,, is considered that includes four wheat disease labels and a healthy wheat label, is used, which was constructed from publicly available datasets. The fine-tuned models are compared for the disease prediction task using the parameters such as accuracy, F1 score, recall and precision. Transfer learning ResNet50 gives promising results once fine-tuned, having the accuracy of 93% compared with other transfer learning models, which ranges from 61 % to 78%. The framework proposed can identify diseases for various crops in agriculture field.