An Improved Deep CNN for Plants Diseases Detection and Diagnosis

Authors

  • Saif Aziz Salman, Bashar Talib Al-Nuaimi

Abstract

Rapid and accurate diagnosis of plant infestation is essential to increase agricultural profitability. In general, human experts are called in to diagnose plant inconsistencies due to disease, anomalies, health deficiencies on the other hand, unusual weather conditions. However, this is expensive, time consuming and sometimes unrealistic. To solve these problems, research into the use of imaging methods for the recognition of plant diseases has become a research domain. In this work, An Automatic System for Diagnosis Plant Disease Based on Deep Convolutional Neural Network (DPD-DCNN) has been proposed. Furthermore, this work distinguishes itself from the previous by employed the DCNN with 12 nested processing layer. It has the ability to detect and diagnose the most common and dangerous types of plant diseases which are Bacterial and fungal viral infections. On the other hand, the most effective machine learning techniques for plant disease detection and diagnosis which are Naive Bayes, and Random Forest have been implemented. Furthermore, this work is proved that the deep learning is most effective in plants disease detection and diagnosis from machine learning. Moreover, this work is implemented by using python programing language and PlantVillage dataset for training and testing the DPD-DCNN system. These images consists of fourteen classes of normal and infected leafs with the total of 22789 images belong to Pepper, Potato and Tomato. However, according to the obtained results it is observed that the DPD-DCNN achieved an excellent results with accuracy of 99.5% during the comparison with Naïve which achieved accuracy of 97% and Random Forest which achieved accuracy of 98%. Also, the performance of the DPD-DCNN has been compared with the related work and it is achieved the highest accuracy.

Published

2020-12-01

Issue

Section

Articles