CNN-Based Rice Nitrogen Status Estimation Using UAV Integrated Multispectral Camera Images

Nitrogen (N) plays an important role in rice growth and productivity. The convolutional neural network (CNN) model already proved its efficiency in N deficiency estimation for rice crops. However, the latest researches still lack the rice N status estimation at an early growth stages. In this study, we devise a rice N status estimation system using the CNN model from the UAV-based multispectral camera images of rice crop at different growth stages. We also perform a comparative study across different CNN model architectures, such as AlexNet, VGG-16, and ResNet-18. The CNN models are trained through two training-testing datasets, which consists of UAV-based multispectral camera-captured field images of two different growth stages of rice crop, tillering and booting. Moreover, an extensive study is carried out by training the CNN models with varying the number of classes related to N applications. Further, the performance of the classifier is measured in terms of accuracy, recall, precision, and F1 score. From the experimental result, we conclude that our proposed system shows better results for the AlexNet model than the other two CNN models. We also conclude that our system shows better validation accuracy with the N applications with three classes than the six classes for the both tillering and booting stage of rice crop.