An automated deep learning based satellite imagery analysis for ecology management

Ecology is the methodical study of biodiversity which affecting natural life and habitats. Due to the anthropogenic pressure on the atmosphere, there is an upraised threat to wild animals and other habitats of the ecological atmosphere. So, there is a need for efficient ecology management models to map and save nature resources. At the same time, the use of satellite imagery analysis is an effective tool for determining important details on earth resources and the platform. It finds useful for proficient ecology management, such as land use detection, forest fire detection, environment planning, and so on. Earlier satellite imagery classification approaches mainly depend upon feature coding approaches which has limited capabilities and yield mediocre outcomes. The recent developments of deep learning models made the image classification highly effective. In this view, this paper presents a new parameter tuned deep learning based EfficientNet model with Variational Autoencoder (PTDLENVAE) model for satellite imagery analysis on ecology management. The presented PTDLEN-VAE model includes a series of operations namely pre-processing, feature extraction, and classification. Primarily, the satellite images are preprocessed to improve the contrast level of the image. Followed by, the PTDLEN based feature extractor is utilized to derive a useful set of feature vectors from the aerial image. Besides, the improved krill herd optimization (IKHO) algorithm is applied for the parameter tuning of the EfficientNet model. Finally, the classification of aerial images using the derived feature vectors takes place by the use of the VAE model. The efficacy of the PTDLEN-VAE model is validated using a benchmark aerial image dataset and the resultant experimental values highlighted the effectiveness of the PTDLEN-VAE model interms of precision, recall, F1score, F2-score, and computation time.