Advanced Identification and Analysis of Forest Canopies in Satellite Imagery Using Deep Learning Algorithms

Abstract

Satellite imagery holds significance across various sectors such as disaster management, law enforcement, and environmental surveillance. These fields necessitate the manual recognition of elements and structures within the images. However, due to vast geographical areas and limited human resources, automation becomes essential. Traditional techniques to perform object detection and classification lack accuracy and dependability for this purpose. Deep learning, a wing of Artificial Intelligence emerged as a promising solution. It has demonstrated efficacy in understanding images through Convolutional neural network[1].Deep Convolutional Neural Networks (DCNNs) have shown outstanding results in most computer vision applications. Traditionally, a fixed data-set is used for detecting the object and other tasks, and after training, model is used exactly as is[2]. In this paper, the data-set consists of pictures of trees and other landscape-related categories that underwent the State-of-the-art CNN algorithm. We extracted a collection of tiles from satellite photos to provide a comprehensive view. And further, we have trained a CNN-based model and have successfully detected the bunch of trees with an accuracy of 95.62 % after fine-tuning. Further, we have trained AlexNet and LeNet models on the same data-set and compared the performance of the proposed model with these two benchmark deep learning models.