Design and performance evaluation of an off-grid hybrid solar system in a remote location: A case study of Legundi Island.

Legundi Island, located in Lampung Bay, Indonesia, has high solar energy potential but still relies on Diesel Power Plants. This research aims to design a hybrid Solar Power Plant and Diesel Power plant system that is environmentally friendly and reliable. Analysis was conducted using the Homer software. •This research designs a hybrid Solar Power Plant system with a capacity of 170.2 kWp and a battery of 1,036 kWh, integrated with three existing Diesel Power Plants units with a capacity of 200 kW.•This system can generate electricity of 296,391 kWh/year with a renewable energy fraction reaching 79.9% and an electricity surplus of 0.988%.•Fuel consumption decreased by 79.3% compared to the existing Diesel Power Plant system.

UNAS-Net: A deep convolutional neural network for predicting Covid-19 severity.

We present a study on Covid-19 detection using deep learning algorithms that help predict and detect Covid-19. Chest X-ray images were used as the input dataset to prepare and train the proposed model. In this context, deep learning architecture (DLA) and optimisation strategies have been proposed and explored to support the automated detection of Covid-19. A model based on a convolutional neural network was proposed to extract features of images for the feature-learning phase. Data augmentation and fine-tuning with deep-feature-based methods were applied to improve the model. Image enhancement and saliency maps were used to enhance visualisation and estimate the disease severity level based on two parameters; degree of opacity and geographic extent. Contrast-limited adaptive histogram equalization and Otsu thresholding were employed with several parameters to investigate the effects on the visualisation results. An experimental investigation was performed between the proposed method and other pretrained DLAs. The proposed work obtained excellent classification accuracy and sensitivity of 97.36% and 95.24% respectively. In addition, the input parameters for image enhancement significantly affected the results. The overall performance metrics were perfect for DenseNet and adequately high for the proposed work which is comparable to other models. Data augmentation and fine-tuning successfully handed the networks to enhance the overall performance, especially in our case with limited datasets.