Utilizing Deep Convolutional Neural Networks for Image-Based Plant Disease Detection

Abstract

Automatic leaf disease detection is a challenging problem in smart agriculture due to the variations in appearance and the complicated backgrounds of plant diseases. Designing a deep convolutional neural network model that extracts visual illness features from the photos and then recognizes the diseases based on the retrieved features is a standard solution to this problem. This method works well when the background is simple, but it has poor accuracy and robustness when the background is complex. The primary factor in identifying leaf-based plant diseases is the color information of sick leaves. Most existing approaches for diagnosing plant diseases depend on the expert diagnosis, which inevitably results in field management and crop disease control behind the times. In this paper, our methodology is based on deep CNN which can be applied to solve these problems to improve the speed and accuracy of disease classification and recognition of plant diseases. We mainly focus on a deep CNN-based model with fine-tuning vegetable leaf diseases dataset. Transfer learning models implemented with famous pre-trained models such as VGG-16 and ResNet-50 are compared with our proposed model. Our proposed model outperforms various existing solutions with an Mean Average Precision (mAP) accuracy of 99.80%.