Abstract

The attribute of image segmentation significantly impacts the validity of the resulting classification, making it an essential step in the image classification process. Present segmentation methods cannot produce a feature set that yields a segmented image of good quality. This work creates a method that yields a set of believable attributes and produces segmented images of excellent quality. The authors aim to create a novel machine-learning model to enhance the image segmentation quality of aerial satellite images using metrics such as Intersection Over Union (IoU), Receiver Operating characteristic (ROC) curves, and accuracy. The Random Forest (RF) algorithm-based machine learning model is intended to separate forested and non-forest regions from aerial satellite images. Finding edges and separating layered objects are two computer vision problems that can be solved using RF, a supervised machine learning model.  Our method objectively assesses the quality of image segmentation by examining the places at which all image objects overlap with the real image regions of a scene item. After generating a collection of features, the RF performed the segmentation process using the Gabor filters and edge detection techniques, such as the Canny and Sobel filters. Segmented images were compared against real masks. The suggested model's superior segmentation capability, with 90\% accuracy, is evaluated against several baseline algorithms, including Linear Discriminant Analysis (LDA), Linear Support Vector Machine (LSVM), and Gaussian Naive Bayes (GNB). For SVM, GNB, and LDA, the corresponding accuracy rates are 81\%, 89\%, and 85\%. 

Statistics