Feature Alignment FPN for Oriented Object Detection in Remote Sensing Images

Abstract

Annotating remote sensing images (RSIs) presents a notable challenge due to its labor-intensive nature. Semisupervised object detection (SSOD) methods tackle this issue by generating pseudo-labels for the unlabeled data, assuming that all classes found in the unlabeled dataset are also represented in the labeled data. However, real-world situations introduce the possibility of out-of-distribution (OOD) samples being mixed with in-distribution (ID) samples within the unlabeled dataset. In this paper, we delve into techniques for conducting SSOD directly on uncurated unlabeled data, which is termed Open-Set Semi-Supervised Object Detection (OSSOD). Our approach commences by employing labeled in-distribution data to dynamically construct a class-wise feature bank (CFB) that captures features specific to each class. Subsequently, we compare the features of predicted object bounding boxes with the corresponding entries in the CFB to calculate OOD scores. We design an adaptive threshold based on the statistical properties of the CFB, allowing us to filter out OOD samples effectively. The effectiveness of our proposed method is substantiated through extensive experiments on two widely used remote sensing object detection datasets: DIOR and DOTA. These experiments showcase the superior performance and efficacy of our approach for OSSOD on RSIs.

Tire defect detection based on low and high-level feature fusion

“…However, the nearest neighbor interpolation upsampling still harms spatial information, resulting in poor network effect in details. Li et al [31] proposed a flow guided upsampling module, which utilizes a novel flow distortion to align features and achieve better cross scale fusion. Park and Paik [32] proposed a pyramid attention upsampling module for object detection by dividing the upsampling process into two branches to extract global contextual information and interpolate and scale feature maps, thereby reducing the loss of semantic information.…”

An Improved TransMVSNet Algorithm for Three-Dimensional Reconstruction in the Unmanned Aerial Vehicle Remote Sensing Domain

“…In this work, we improve the TransMVSNet network, which uses a feature extraction module (FPN) [31] as its feature extraction network. In this paper, we use an asymptotic feature pyramid network (AFPN) [32] to support the direct interaction of non-adjacent layers.…”

A Novel Remote Sensing Spatiotemporal Data Fusion Framework Based on the Combination of Deep-Learning Downscaling and Traditional Fusion Algorithm

“…In conventional fusion methods, this leads to poor predictions of the boundaries between features. The accuracy of spatiotemporal data fusion depends largely on the accuracy of the input data, so an improved or innovative input data downscaling method can improve the performance of most spatio-temporal data fusion methods; for example, Zhai et al [21] and Xie et al [22] produced good results, and Li et al [23] proposed a MODIS strip noise cancellation strategy for spatio-temporal fusion methods, which better improves the accuracy of the fusion results. In addition, we found that compared to traditional spatio-temporal fusion methods, the deep learningbased spatio-temporal fusion algorithms bring more noise to the image fusion process and ignore the advantages of traditional spatio-temporal fusion algorithms in terms of image element unmixing and land use type change prediction.…”