Implementing CNN and RF Models for Multi-Level Classification: Deciphering Beetroot Aphid Disease Severity

Beetroot aphid infestation is one of the most serious threats to agricultural yield and food security. In this study, we hope to use machine learning models based on Convolutional Neural Networks (CNN) and Random Forest (RF), among others, that can accurately classify the three different severity levels of beetroot aphid disease. A complete dataset was assembled, containing 14,000 images across six levels of severity, and carefully preprocessed for network training. The developed models achieved outstanding results, with an overall accuracy of 97.92 % in identifying levels of severity. The CNN, an image classification expert itself, distinguished very accurately and precisely the complex patterns in these images. Meanwhile, the RF model, using feature extraction techniques, was very accurate and there were differences between levels of severity. By comparing these models, the relative performance subtleties and possible practical applications can be seen. Both models show very high levels of accuracy, but issues resulting from practical considerations suggest that CNN should be used in scenarios where it is important to have visual clues. Meanwhile, assuming limited resources, RF may play a bigger role. These study results greatly facilitate applications of precision agriculture since they create common tools that can be used to make more timely and accurate evaluations of disease severity--information that is essential for effective crop protection plans and operations.