The Smart Agriculture based on Reconstructed Thermal Image

Abstract

The utilization of thermal image in supporting precision agriculture is tremendous nowadays. There are many applications of thermal images in agricultural fields, such as detecting crop water stress, monitoring of free-range rabbits, measuring of crop canopy temperature and so on. Furthermore, the importance of thermal camera became the urgent need of perform the smart agriculture. Otherwise, the price of thermal camera is very expensive todays. Then, this kind of camera is not easy to find in the market. Therefore, it makes the utilization of implementation thermal images difficult. In order to handle this problem, the proposed method intends to generate thermal image from visible images. Further, the thermal information concerning with the agriculture, especially the fertility of leaves in paddy fields and the water stress can be monitored. The proposed method uses deep learning architecture to learn the thermal and visible image dataset. It applies Generative Adversarial Network architecture. This GAN pre-trained model trained using 150 images of training dataset and tested using many images of testset. The obtained model is used for generating thermal images from visible images. The results show the constructed thermal image has high accuracy. The assessment metric uses SSIM and PSNR methods. Their indexes show that the results have the high accuracy. The visual assessment shows the reconstructed thermal images also have high precision. Finally, the constructed thermal images can be implemented in smart agriculture purposes.