AUTOMATED FEATURE BASED DETECTION AND PREDICTION METHODS FOR RICE DISEASES USING DEEP NEURAL NETWORKS
Keywords:
CNN, CNN models, deep learning, transfer learning, disease detection, classification, fine tuningAbstract
Crop diseases have serious economic impacts if not detected at an early stage. Rice is an important cash crop in Asia and is the backbone of the economy of many countries. Therefore, accurate and rapid detection of plant dis- eases is critical, but a difficult task. Brown spots, leafblast and Hispa are common diseases of rice plants that are visible to the human eye. However, it takes time to manually inspect/detect a large field and cannot prevent the rapid spread of the disease.Deep learning provides accurate, fast, and independent detection and recognition through image classification.This re- search work proposes Deep Learning algorithms to detect rice diseases. A rice dataset used in this work is based on four labelled rice classes out of which three (Brownspot, Leafblast, Hispa) are diseased and one is healthy (having healthy leaf images). This research work is divided into two parts, A comparative analysis of pre-trained deep learning models which comprises VGG16, VGG19, InceptionV3 and ResNet50. VGG16 reached the highest accuracy 94% of the benchmarking. secondly, two models Stacked Convolve Neural Network (SCNN) and Dense Classified Neural Network (DCNN) are offered for detection and classification purposes. SCNN achieves 98% preci- sion while DCNN achieves 99.96% drive precision.
