Paper Title
Tobacco Leaf Disease Diagnosis System Using Vision Inference and Deep Convolutional Neural Networks for Small-scale Farmers in Ilocos Sur

Abstract
This study presents the development and evaluation of a tobacco leaf disease diagnosis system that integrates object detection and deep learning techniques. The prototype, equipped with a camera, is designed to classify the conditions of tobacco leaves through either digital images or real-time input from the camera. The system employs a YOLO model for initial object identification and healthy leaf detection, followed by a deep convolutional neural network (DCNN) to classify diseased leaves into three categories: frogeye spot, tobacco mosaic virus (TMV), and wildfire. To validate the system's performance, two testing approaches were employed: digital image testing, which eliminated environmental factors, and controlled physical testing using actual leaves under varied illumination and camera heights. The results indicated that the system achieved a mean classification accuracy of 94.44% during digital testing, with a low standard deviation, reflecting consistent performance. In the prototype testing phase, statistical analysis using two-factor ANOVA with replication revealed that both illumination and camera height significantly impacted accuracy, with a notable interaction between these two factors. The best performance was observed at a height of 33 cm under illumination of 10.83 lux, achieving 91.67% accuracy. The system's real-time functionality and high accuracy underscore its potential as a practical tool for automated plant disease identification in enclosed or controlled agricultural environments. Keywords - DCNN, Raspberry Pi, Tobacco leaf disease detection, vision inference, YOLO