Global Multi-Scale Information Fusion for Multi-Class Object Counting in Remote Sensing Images

Abstract

In recent years, object counting has been investigated and has made significant progress under a surveillance-view. However, there exists only a few works focusing on the remote sensing object density estimation, and the performance of existing methods is not promising. On the one hand, due to the imbalance distribution of targets in remote sensing images, the model might collapse, leading a severe degradation. On the other hand, the scale of targets in remote sensing images actually varies in real scenarios, which remains a challenge for counting objects accurately. To remedy the above problems, we propose an approach named “SwinCounter” for object counting in remote sensing. Moreover, we introduce a Balanced MSE Loss to pay more attention to the fewer samples, which alleviates the problem of imbalanced object labels. In addition, the attention mechanism in our SwinCounter can precisely capture multi-scale information. Thus, the model is aware of different scales of objects, which capture small and dense targetes more precisely. We build experiments on the RSOC dataset, achieving MAEs of 7.2, 151.5, 14.38, and 52.88 and MSEs of 10.1, 436.0, 22.7, and 74.82 on the Building, Small-Vehicle, Large-Vehicle, and Ship sub-datasets, which demonstrates the competitiveness and superiority of the proposed method.

Feature Balance for Fine-Grained Object Classification in Aerial Images

“…To provide comprehensive benchmarks, we evaluate some popular object-counting methods on the NWPU-MOC dataset. These methods include the crowd-counting method (MCNN [24], CSRNet [39], SFCN [52], SCAR [53]) and the objectcounting methods designed for remote sensing scenes (ASP-Net [14], PSGCNet [15], SwinCounter [50]), respectively. The experimental results on the NWPU-MOC dataset, where two types of Gaussian kernels are used to generate Gaussian density maps, are presented in Table II.…”

“…and F 3 ∈ R 4C×H/16×W/16 (C=128), respectively. To alleviate the scale variation problem present in aerial scenes, following the previous work [50], we fuse multi-scale feature information by using a feature pyramid network(FPN),…”

“…Following the previous works [14], [50], the MSE loss function is used as the counting loss. However, for the MOC task, different from single-category object counting, there could be overlaps in the same spatial locations of each channel of predicted density maps.…”

Object counting in remote sensing via selective spatial‐frequency pyramid network

“…Firstly, by performing object counting at the edge, the computational burden on centralized cloud servers is reduced, which improves efficiency and responsiveness. 5 This allows for real-time processing of remote sensing imagery, which is particularly critical in applications requiring timely and accurate information, such as environment monitoring, [6][7][8] urban planning, 9,10 and disaster management. 11 Secondly, the integration enhances data security and privacy protection.…”