Comparison between Alexnet, Googlenet and ResNet-50 for brownspot, hispa and leafblast disease classification / Siti Maisarah Zainorzuli

• Main Author: Zainorzuli, Siti Maisarah
• Format: Thesis
• Language: English
• Published: 2023
• Subjects: SB Plant culture; Pests and diseases
• Online Access: https://ir.uitm.edu.my/id/eprint/91113/1/91113.pdf; https://ir.uitm.edu.my/id/eprint/91113/
• Institution: Universiti Teknologi Mara

Rice is a staple food for Malaysian, hence ensuring its production is essential. There is a large manufacturer of rice in Malaysia every year to contain the need of millions of Malaysians but still not sufficient. There are a number of causes for the decrease in production, such as poor management of pests and diseases, fertiliser, and shortage of specialists. In Malaysia, paddy fields are infected by diseases and pest known as brown planthopper (BPH) which could potentially reduce the yield by about 30%-50% and 15%. The conventional method, a paddy disease specialist is required to identify and diagnose the paddy leaf disease. The paddy disease expert will obtain several samples of paddy leaf images from the farmer. Afterwards, the required sample was sent to the biotech laboratory so that the affected leaf can be analysed. The process for this method was time-consuming, inconvenient for the farmer and it was very costly. Since the 1990s, the use of computers and information technology has enhanced the methods of agriculture and been beneficial in the agricultural industry. Thus, with the application of Deep Learning method the diseases can be detected at the early stage. Precaution measures can be taken to lessen the damage as soon as possible. The objective of this work is to classify the types of paddy disease such as brownspot, leafblast and hispa by using CNN model. Thus, this research conducts an analysis using several types of CNN to validate and compare the performance of proposed model in terms of precision, recall, F1 score, and robustness with AlexNet and GoogleNet. Hence, the result shows ResNet-50 is exceptional in all indexes except time by using the optimal configuration namely learning rate at 0.001 and number of epochs at 30. The time of ResNet-50 (1626s) is close to AlexNet but worse than GoogleNet. For each index, ResNet-50 above 90%. Meanwhile AlexNet and GoogleNet only obtained 94.84% and 95% for accuracy, while the rest only obtained around 88% to 89%. All indexes of ResNet-50 are the first ranked then followed by GoogleNet. In short, ResNet-50 is a more accurate and precise model than others.