Journal of Management Information and Decision Sciences Volume 24, Special Issue 6, 2021 1 1532-5806-24-S6-67 Citation Information: Aldakheel, E.A., & Zakariah, M. (2021). Detection of targeted region using deep learning-based multiscale alexnet cnn scheme for hyperspectral satellite image. Journal of Management Information and Decision Sciences, 24(S6), 1-15. DETECTION OF TARGETED REGION USING DEEP LEARNING-BASED MULTISCALE ALEXNET CNN SCHEME FOR HYPERSPECTRAL SATELLITE IMAGE Eman Abdullah Aldakheel, Princess Nourah Bint Abdulrahman University Mohammed Zakariah, King Saud University Riyadh ABSTRACT In remote sensing, segmentation and classification of satellite images are a demanding task that attributes various kinds of detecting the targeted regions. The accurate segmentation of the targeted area is considered a crucial portion of seeing several geographical positions and locations. In turn, the classification and segmentation technique enhances the detection rate and accurately recognizes the target regions with reduced execution time. A deep learning- dependent automatic detection, segmentation, and classification of satellite images are conducted in this process, employing artificial intelligence methods. At first, the input image is preprocessed, segmented using semantic-ROI segmentation, and classified using the Multiscale Alex Net CNN classifier method. The semantic-based ROI segmentation stage is employed for extracting the regions. By using a multilinear spectral decomposition-based extraction approach, the spatial information is removed. Then the deep learning-based Multiscale Alex Net classifier effectively classifies the satellite images. The dataset employed are Indian Pines, Pavia, and Salinas. The presented approach performance is estimated and the outcomes attained are evaluated to prove the efficiency of the proposed system. Keywords: Remote Sensing, Satellite Images, Multi-Scale Alex Net CNN Classifier, Semantic- Based ROI Segmentation, Multilinear Spectral Decomposition-Based Extraction INTRODUCTION Nowadays, researchers have concentrated on classifying images worldwide, which is a powerful technology for the pattern recognition and computer vision field. Typically, satellite images in digital form are processed to attain the target regions (Zhang, 2017). A technique of image processing was employed for regaining the precise data set from images that were kept. Therefore, this method visually improved the perception and intended for repairing or modifying the image, which depends on the blurring, image deformation, or image deterioration (Chen, 2017). Several techniques were available for image analysis, and they rely on some desired problem's requirements (Huang, 2020; Kamir, 2020; Avolio, 2017). The image classification and segmentation algorithms were conducted in many thematic groups of image regions. Image classification forms are characterized as quarter sets in dissimilar viewpoints similar to clustering of the image depending on edge-based, region, and model-driven (Soldin, 2018). Because of this modest code alike quick and robust effect of classification, CNN is widely employed. Furthermore, the images are generally classified through a CNN on it's own; it is feasible for getting fixed in local maximum, the boundaries blurring through bad visual outcome. The primary problems are capturing images yet similarly to process the captured information quickly and promptly broadcast them at the detailed detection of a target, and their accuracy assesses the classification. The technique for a problem associated with the target recognition is to separate several scene components and distinguish one fascinating one from the enduring. Still, the classification approach is a complex task and a legal concern for detecting robust satellite target systems. Moreover, there will be some problems in recognizing the guards