Improved Feature Extraction and Object Detection Accuracy with the Novel DenseNet Algorithm Compared to the SqueezeNet Algorithm in Remote Sensing Images

As the demand for accurate object detection systems increases across various fields, understanding the strengths and weaknesses of different algorithms becomes crucial. Despite numerous advancements in deep learning, there remains a lack of comprehensive comparisons between lightweight models like SqueezeNet and more complex architectures such as DenseNet. Previous research has primarily concentrated on isolated performance metrics without addressing the comparative effectiveness in real-world applications. This study's main focus is evaluating and contrasting SqueezeNet and DenseNet's performance in feature extraction and object detection tasks. Within the domain of our Deep Learning analysis, 48 samples are analyzed in the dataset. Two sets of these samples were created in order to conduct a thorough analysis. Group 1 used the DenseNet model, which has 24 iterations, while Group 2 used SqueezeNet, which has 24 iterations as well. Included in the dataset were 324 jpg images related to item detection and 724 jpg images showing different kinds of vegetation. The results demonstrate the higher performance of DenseNet, with an accuracy of 79.67% vs SqueezeNet's 61.21%, when comparing the efficacy of the two algorithms in object detection. Therefore, with a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{p}$</tex>-value of 0.254 (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$p&lt;0.05$</tex>), no statistically significant difference was found between the two groups. This research, which covers both SqueezeNet and deep learning, unequivocally shows that DenseNet performs significantly better than SqueezeNet in the crucial object detection task.