Semantic segmentation of high resolution imagery using U-Net for classification of sugarcane age in parts of Lumajang Jember Regency

Abstract Sugar cane mapping at a lower cost and time is an important need for farmers and the sugar industry. Semantic image segmentation with U-Net offers a potential solution to overcome this problem. This research uses high resolution Google Satellite imagery to examine the area around the Jatiroto Sugar Factory in Jember Regency, East Java Province. The aim of this research is to carry out semantic segmentation on high resolution images using U-Net to determine the age classification of sugarcane. The results showed that U-Net was able to achieve sugarcane age classification accuracy of 83.6%. Disturbances in the form of differences in recording and contrast of image mosaic results result in the learning curve experiencing overfitting. Image segmentation reveals the presence of new objects beyond the training data, necessitating future development related to the findings of visual interpretation, field data, and segmentation data. In conclusion, U-Net shows promising potential for sugarcane age classification in Google Satellite imagery. Further research is needed to improve accuracy and overcome the various obstacles encountered.