Interpretation of Transplanted Positions Based on Image Super-Resolution Approaches for Rice Paddies

Abstract

Due to rapid developments in aerial photography techniques, drones are now capable of providing essential, full-color images for rice paddy field applications. In this article, a technique is introduced that employs an unsupervised model based on generative adversarial networks and an image super-resolution approach to increase the resolution of full-color images acquired by drones. These improved images are then utilized to detect and interpret the locations of transplanted rice paddies. The process involves the use of advanced image processing techniques to enhance the clarity and detail of drone images. Validation was conducted using an 80/20 training and testing data ratio, and a set of established rice paddy seedling coordinates was used to assess the effectiveness of the model. Based on the obtained results, the accuracy rate for identifying and interpreting the transplanted positions in rice paddies is demonstrated to be above 93%, as measured by the F1-measure value.