Color Recognition of Rice Plant Leaves to Monitor Rice Plant Growth with Drones using Deep Learning
Mochammad Zen Samsono Hadi- Syifa Aulia Widya Ananda- Mike Yuliana

Politeknik Elektronika Negeri Surabaya

Abstract

Rice is an important crop as a staple food source in Indonesia and even around the world. The vast rice farmlands in Indonesia are present to fulfill the need for rice in this country and even in foreign countries. However, in rice farming, there are three growth phases that must always be observed by rice farmers for a successful harvest, namely the vegetative, generative, and ripening phases. The vast amount of agricultural land owned by farmers, especially those who have more than one land, is certainly very labor-intensive if they have to carry out regular monitoring every day. Therefore, in this research, an automatic system is made to monitor the growth of rice plants using the Deep Learning MobileNet algorithm. This system will classify aerial images that have been taken using drones into four color classes according to the leaf color chart. The total dataset will be divided into 3 parts for training, validation, and testing with a ratio of 70- 20- 10. The testing is done with five test scenarios by changing the learning rate parameters. There is a decrease in loss with every change in learning rate. It can be observed that the learning rate of 0.005 has the lowest loss value and has the highest AP and AR values. The more the number of iterations or num step, the loss value is closer to zero. The iteration 30000 at learning rate 0.005 has the lowest loss value.

Keywords: Deep Learning, MobileNet, monitoring rice plant growth

Topic: Green Infrastructure