Transforming Sugarcane Leaf Diseases Pathology with Convolutional Neural Networks and SVM
Kapil Rajput, Manika Manwal, Rajeev Kumar Chauhan, Vinay Kukreja, Shiva Mehta
Abstract
This paper has provided, in detail, a complete study of the application of Convolutional Neural Networks (CNN) together with support vector machines for the classification of multiple classes of sugarcane leaf disease In sugarcane cultivation, where early and accurate diagnosis of diseases is considered essential for optimal agricultural productivity, the study describes a novel AI-based method to detect and classify leaf disease that aims at improving crop health management. Each disease class was evaluated using precision, recall, F1-score support proportion and accuracy metrics to assess the model's performance. The performance metrics were encouraging, with the precision reaching 73.24%, recall of 77.20% and F1-score at 65.19% for Class sample size = (no(t) x no(k))/(total number), where not is No_of test class no(). It can be noted that CKSAAP+ proved to yield. The support and the proportion of supporting votes for these classes ranged between 851 and 114, respectively, whereas the overall accuracy remained high, ranging from 88% to 92%. Further analysis provided an exciting view of the system using macro, micro and weighted averages. The macro averages were 75.25% for precision, 75.51% for recall, and balanced performance across all classes with an F-measure of the value equal to 74%. Considering the class distribution, the weighted averages resulted in slightly improved precision of 75.79%, recall - 75.39%, and F1-score - 0.48%. The micro averages, which measure the total true positives, false negatives and false positives across all classes, scored 75.4 % for precision-recall and F1 score, highlighting the model's performance stability.