<![CDATA[This research was conducted to mapping the boundaries of paddy fields in Kepanjen District, Malang Regency, using the U-Net deep learning model based on Convolutional Neural Network (CNN) for image classification per pixel. The input used was WorldView-2 imagery, and the model was trained with 25, 75, and 150 epochs variety to evaluate its performance in accurately classifying paddy field use down to the parcel scale. The 75 epochs variation model was used for image classification because it has balance between model performance and training period. Validation test were conducted using paired T-test to identify statisfically significant differences between the classified image paddy area and the actual field conditions. The results showed that U-Net model with various epoch variations did not differ significantly in terms of performance, the time taken for the model to learn the dataset per epoch, or accuracy, which reached 90%. The model was able to accurately classify paddy and non-paddy land use in WorldView-2 imagery down to the plot scale. Based on WorldView-2 image segmentation to identify paddy field boundaries, the paddy field area is 1217 Ha. The results were validated by comparing the area of rice fields in the image with the actual area of rice fields in the field. Validation test result with actual paddy area showed a calculated T (0,5486) lower than the T table (1,9603) and a p value (0,5833) greather than 0,05. This indicates no significant mean difference between the two data sets (ground check and imagery). The lack of significant difference provides strong evidence that the U-Net model is effective for mapping paddy plot boundaries on a large scale.]]>
