Predicting Chlorophyll Intensity Of Various Plants Using Improved Convolutional Neural Network

Michelle, Nashrin Bawai (2023) Predicting Chlorophyll Intensity Of Various Plants Using Improved Convolutional Neural Network. [Final Year Project Report] (Unpublished)

	PDF MICHELLE NASHRIN (24pgs).pdf Download (724kB)
	PDF (Please get the password by email to repository@unimas.my , or call ext: 3914 / 3942 / 3933) MICHELLE NASHRIN (Fulltext).pdf Restricted to Registered users only Download (3MB)

Abstract

Chlorophyll pigment is beneficial during photosynthesis to absorb sufficient light energy and provide nutrients to the plant to grow healthy. A convenient assessment of chlorophyll content is essential in smart management agriculture. Several attempts have been made to implements computer vision to enhance the precision agriculture techniques. However, problems still arise as the applied algorithms of machine learning are time consuming, very complex architectures, high computational cost, as well as less generalization towards various plant species and size of datasets. Hence, in this project, a rapid and straightforward convolutional neural network (CNN) algorithm was proposed to predict chlorophyll intensity of various plant species based on leaf reflectance spectra. The datasets were taken from ANGERS Leaf Optical Properties Database (2003). The proposed model consists of Hybrid CNN as a feature extractor and support vector regression (SVR) network as a predictor. Hybrid CNN was designed by modifying the architectures of AlexNet and PNet using MATLAB R2023a. The performance of Hybrid CNN with SVR (CNN-SVR) was also compared with AlexNet, PNet, and SVR. Results showed that the best CNN also can be designed with one input, four convolutional, four max-pooling and three fully connected layers which can be found in Hybrid CNN-SVR. The experimental results show that the prediction accuracy of chlorophyll intensity is satisfying with a mean square error (MSE) of 0.1558 and 1.149 for training and testing sets, respectively.

Actions (For repository members only: login required)

View Item