A Fuzzy logic Clustering Classification of Satellite Images Using K-Means Clustering

In this study, a unique strategy to improving the semantic segmentation of satellite photographs with complicated backdrops is presented. This approach makes use of deep learning techniques, specifically an algorithm called a K-means clustering algorithm. Satellite imaging is an extremely helpful instrument that can be used to gain a current and up-to-date portrayal of the ecosystem that the Earth and its surroundings are a part of. By continually recording photographs in real time, the vast network of remote sensing satellites that circle the Earth provides coverage that is both comprehensive and consistent over the whole planet. This feature makes it possible to implement a broad variety of applications that are beneficial to the human race. Significant interest in a variety of applications has been sparked as a result of the increasing accessibility of routinely collected satellite photographs of a variety of places. Image segmentation is an extremely important technique that is utilized in the field of satellite image processing. This technique is utilized for the purpose of extracting and compiling vast amounts of data from satellite photographs. Because of its potential to derive and evaluate information from a picture, the process of extracting and segmenting satellite images has been the subject of a great deal of research. Image segmentation is an essential stage in the field of satellite image processing because it simplifies the process of extracting and compiling vast amounts of information from satellite photographs. There is a paucity of self-supervised deep learning algorithms that hold the capacity to partition an image into a specified number of separate clusters, such as the k-means algorithm, while simultaneously applying an end-to-end deep learning approach at this time. One example of such an algorithm is the k-means algorithm. In this part, we will discuss the k-Means technique, which permits unsupervised segmentation of a 4-band picture (a picture that contains the near-infrared, red, green, and blue bands), into k separate classes. Our method demonstrates that it is possible to do discrete search using convolutional neural networks (CNN) in conjunction with gradient descent software. [CNN] stands for "convolutional neural networks." Within the model, the model is able to identify k different clustering classes, each of which is represented by a discrete binary mask and its matching texture that was developed independently. When put together, these parts provide an accurate representation of the original picture. By combining the concepts of fuzzy logic with the idea of membership function, the purpose of this investigation is to develop a solution to the problems caused by the traditional implementation of the k-means algorithm. When evaluating the efficacy of the method, several different cluster validity metrics are taken into consideration and the accuracy is about at 87.25 percent and 97.75 percent.