Prediction of Plant Diseases using Simple Novel Image Detection Technique with Improved Accuracy and Compared with Convolutional Neural Network

Abstract

Aim: The aim of this research is to compare the performance of Simple Novel Image Detection Technique (SNIDT) with the Convolutional Neural Network (CNN) to improve the accuracy of Plant Disease Detection for Plant Protection.
Materials and Methods: The implementation of research is done using SNIDT, is a Machine Learning algorithm which calculates the RGB pixel density using the input images of plant leaves which are affected by diseases. The obtained pixel density is used for the summation of pixel density, the image gets converted to gray scale and then the normalization and brightness are measured. The algorithm classifies the output based on the grayscale of the given input image. Simple Novel Image Detection Technique and Convolutional Neural Network performance was compared with the samples of N=50 per group during the implementation. The research work uses 100 sample images for comparing both algorithms' performance. The sample size was calculated using the G Power statistical tool using G power with Pretest power 0.8.
Results: The comparison shows that Simple Novel Image Detection Technique has better mean accuracy of 81.13%, compared with Convolutional Neural Network the mean accuracy produces 75.09% with the significant value of .025 for p<0.05 provides significance.
Conclusion: The results show that the Simple Novel Image Detection technique has better accuracy than the Convolutional Neural Network algorithm and has significance.