Counting the uncountable: deep semantic density estimation from Space

TL;DR

This paper introduces a novel deep learning approach combining density estimation and semantic segmentation to count small, occluded objects in satellite images, addressing challenges posed by their size and scene clutter.

Contribution

It presents an end-to-end CNN model that effectively counts and distinguishes small objects in remote sensing images, a task difficult for traditional detection methods.

Findings

Deep semantic density estimation robustly counts objects in cluttered scenes.

Specific CNN architectures are necessary for remote sensing tasks.

The method outperforms traditional detection approaches for small object counting.

Abstract

We propose a new method to count objects of specific categories that are significantly smaller than the ground sampling distance of a satellite image. This task is hard due to the cluttered nature of scenes where different object categories occur. Target objects can be partially occluded, vary in appearance within the same class and look alike to different categories. Since traditional object detection is infeasible due to the small size of objects with respect to the pixel size, we cast object counting as a density estimation problem. To distinguish objects of different classes, our approach combines density estimation with semantic segmentation in an end-to-end learnable convolutional neural network (CNN). Experiments show that deep semantic density estimation can robustly count objects of various classes in cluttered scenes. Experiments also suggest that we need specific CNN architectures in remote sensing instead of blindly applying existing ones from computer vision.