Multiclass aerial image recognition using improved Black Widow Optimization with deep learning on unmanned aerial networks imaging

Abstract

Unmanned aerial vehicles (UAVs) have transformed industries by capturing high-resolution aerial images with specialized cameras and sensors. This technology is widely used in remote sensing (RS), environmental monitoring, disaster response, agriculture, and urban development. UAV-based aerial image classification improves the analysis of large-scale geographical and environmental data. Land cover classification (LCC) is a crucial application of UAV imaging, involving the categorization of land surfaces like vegetation, water bodies, urban areas, and bare soil. It offers valuable insights for environmental monitoring, urban planning, and resource management. Conventional methods for image classification often encounter challenges in handling the complexity and scale of UAV-generated data. Deep learning (DL), particularly convolutional neural networks (CNNs), has become a crucial tool for precisely classifying aerial images. DL uses neural networks to improve land cover recognition, simplify object detection, and enable real-time data analysis. This study proposes a Multiclass Aerial Image Recognition using Improved Black Widow Optimization with Deep Learning (MAIR-IBWODL) approach to UAV imaging. The main intention of the MAIR-IBWODL approach is to identify and categorize numerous classes that occur in the images. To attain this, the MAIR-IBWODL method utilizes the SE-DenseNet method for learning complex feature patterns from RS images. Furthermore, the MAIR-IBWODL method employs the IBWO method for the hyperparameter range of the SE-DenseNet model. Also, the attention long short-term memory (ALSTM) technique is implemented for classification. To elucidate the performance of the MAIR-IBWODL technique, a sequence of simulations is performed under the UCM Landuse dataset. The experimentation validation of the MAIR-IBWODL technique depicted a superior accuracy value of 99.94% over existing models.