Disease detection system in rice leaves, using Deep Learning models (#479)

Authors

Villegas-Cubas, Juan Elias

Valdiviezo-Sandoval, Hector David

Abstract

This research focuses on the issue of diseases in rice cultivation, a critical concern for the Peruvian economy. Early detection of these diseases is vital to mitigate adverse impacts on production. The proposed model was trained using the "Paddy Doctor" dataset, comprising 10,407 RGB images with a resolution of 480 x 640 pixels, categorized into 10 classes. A comparison between a CNN and InceptionV3 revealed InceptionV3's significant superiority in recall (74.107% vs. 46.923%) and precision (87.680% vs. 80.947%) after 100 epochs. For the final evaluation of the InceptionV3 model fine-tuned with 400 epochs, results included a accuracy of 88.00%, precision of 90.31%, and recall of 86.10%. Despite an 88% accuracy, the loss degree suggests potential enhancements, such as augmenting the dataset with more images. The detection system, developed using TensorflowJS and Angular, is accessible online, offering a valuable tool for farmers and promoting enhanced production and sustainability in the rice sector