Semantic Segmentation of Rice Leaf Blast Disease using Optimized U-Net

Rice is one of the staple crops for primary food needs in Indonesia. In the rice sector, pest and disease control is essential in maximizing agricultural production potential because several cases cause a decrease in rice yields due to pests or diseases. The need for checking as anticipation of pests in rice is a concern in this study. This research was conducted with three different optimizations on several U-Net parameters for semantic segmentation, and the optimizations used include Hyperband, Random Search, and Bayesian. The results shown from the three optimization experiments show different efficiency. From our results, Random search has attained several achievements with the smallest number of parameters and the highest accuracy at 8,6 million and 98.5%, respectively. In contrast, the lowest loss value was obtained by Hyperband optimization with a value of 0.0433 each per 50 epochs. In future work, the measurement of the diseases is required.