Paddy Plant (Oryza Sativa) Disease Detection Based On Improved Convolution Neural Network

• Main Author: Marceila Suzie, Ambrose
• Format: Final Year Project Report
• Language: English; English
• Published: Universiti Malaysia Sarawak, (UNIMAS) 2023
• Subjects: GE Environmental Sciences; T Technology (General)
• Online Access: http://ir.unimas.my/id/eprint/43109/1/Marceila%20Suzie%2024%20pgs.pdf; http://ir.unimas.my/id/eprint/43109/2/Marceila%20Suzie%20ft.pdf; http://ir.unimas.my/id/eprint/43109/

Paddy plant (Oryza sativa L.) plays a pivotal role in ensuring food security in Malaysia and enhancing the agricultural ecological environment. The early detection of paddy plant diseases is very important in the agriculture sector. It helps to pave the way for effective decision-making in overcoming the issues caused by the paddy diseases. Some existing methods, for example, GoogleNet, InceptionNet and VGGNet are used to detect paddy diseases. However, the applied algorithms of machine learning have some drawbacks in terms of time consuming, complex architecture, high computational cost, and the ability of the model to perform with more various types of paddy disease. Therefore, in this project, a hybrid model based on an improved Convolution Neural Network is developed to detect paddy diseases. The combination of the model is between DenseNet-201 and CAE network, and this hybrid model known as CAE-DenseNet. The model is designed, trained, and tested by using MATLAB 2023a. One of the purposes of the combination is to improve the feature extraction technique. Then, by using the large data size with the application of data augmentation, the performance of the model improved. The performance of proposed model CAE-DenseNet is compared with other existing models such as original CAE network, AlexNet, GoogleNet, ResNet-50, VGG- 16 and original DenseNet-201. The results analysis shows that the performance of existing models is average because of their limitations to train the data and detected the paddy diseases. The performance of CAE-DenseNet model also evaluated with cross validation and confusion matrix. Overall, the proposed model CAE-DenseNet has a high detection accuracy rate of 88.15%. The accuracy result based on 10-K fold cross validation is 91.11% and the value for precision, recall and F1 score from confusion matrix are 92.03%, 84.70% and 88.21% respectively. Therefore, it proved that the hybrid model CAE-DenseNet able to detect the paddy diseases efficiently.