LENET-CNN FOR FEATURE EXTRACTION AND ACCURATE CLASSIFICATION OF REMOTE SENSING MULTI-SPECTRAL SATELLITE IMAGES USING MACHINE LEARNING

Abstract

For a variety of uses, such as disaster relief, engineering, law enforcement, and environmental monitoring, satellite imagingis strongand indispensable. Con- ventional techniquesfor object detection and categorization are imprecise and unreliable. Applications that automatically classify and detect items in photos are needed to address this problem. Image processing is performed to extract or enhance images with high quality. This study suggested Machine Learning (ML) techniques that improve the classification of remote sensing photos and identify objects to make the images more convenient. A subset of artificial intelligence (AI) called machine learning uses information from past events to learn and adapt how a computer program behaves. Color and image pixel values are grouped into consequential categories as part of the multi-spectral satellite image categoriza- tion process. Accurate satellite image classification and feature extraction from remote sensing images is accomplished using a hybrid approach that combines the LeNet and Convolution Neural Network (CNN) algorithms. The concept of a neural network to recognize patterns much like human brains is employed. The proposed system uses a hybrid algorithm to achieve feature extraction and object detection through machine learning techniques .The dataset includes a variety of large-scale photosand collections of satellite images gathered from the Kaggle database, which is the largest community in the world offering a data platform for learning data science. Color, Texture, and structure are the three main cat- egories from which the hybrid techniques are employed to extract the feature. The best optimizer of LeNet helps separate hyper spectral images based on their shape and texture. CNN can classify land cover types with higher accuracy, es- peciallyin mining areas. CNN function analyse data and recognize patterns for accurate classification of images.Semantic segmentation of a remotely sensed image in the temporal, spatial, and spectral-domain is an essential pre-processing step where different classes of objects like water bodies, industrial, crops, pas- ture, and roads are localizing by a boundary. The results of the inquiry show that the suggested Hybrid LeNet and Convolution Neural Network (CNN) algorithm provides high-accuracy object detection and feature extraction through machine learning techniques.