Semantic-Edge-Supervised Single-Stage Detector for Oriented Object Detection in Remote Sensing Imagery

Abstract

In recent years, significant progress has been made in arbitrary-oriented object detection. Different from natural images, object detection in aerial images remains its problems and challenges. Current feature enhancement strategies in this field mainly focus on enhancing the local critical response of the target while ignoring the target’s contextual information, which is indispensable for detecting remote sensing targets in complex backgrounds. In this paper, we innovatively combine semantic edge detection with arbitrary-oriented object detection and propose a feature enhancement network base on a semantic edge supervision module (SES) that realizes an attention-like mechanism in three dimensions of space, channel, and pyramid level. It helps the network pay attention to the edge features of targets at multiple scales to obtain more regression clues. Furthermore, to solve the problem of dense objects with different directions in remote sensing images, we propose a rotation-invariant spatial pooling pyramid (RISPP) to extract the features of objects from multiple orientations. Based on the two feature enhancement modules, we named the network SE2-Det; extensive experiments on large public datasets of aerial images (DOTA and UCAS-AOD) validate our approach’s effectiveness and demonstrate our detector’s superior performance.

A multiple directions turning vehicle counting method at intersections based on arbitrary-oriented detection and stack Long Short-Term Memory

“…Zheng et al [16] proposed Adaptive Dynamic Refined Single-stage Transformer Detector to improve their methods for better oriented object detection in remote sensing images. Cao et al [17] combine semantic edge detection with arbitrary-oriented object detection to obtain more regression clues.…”

“…The vehicles passed by without being counted successfully mainly due to the failure to be detected and the misclassification of sLSTM. It can be seen that the proposed multi-turning counting model based on sLSTM can achieve accuracy Faster RCNN [29] 88.79% 88.89% Retina Net [30] 87.23% 86.79% Gliding-Vertex [31] 88.77% 88.79% Heat-RCNN [32] 90.66% 90.69% CBNet [33] 90.36% 90.11% SESNet [17] 90.56% 90.41% Kfiou [34] 90.63% 90.59% Ours (SCAMNet) 90.92% 90.87% values of 96.77%, 97.92%, 98.31%, 100%, 97.83%, 94.87%, 97.75% and 95.38% on eight different turnings respectively. The average accuracy is 98.18%.…”

“…The methods of Faster-RCNN [29], Retinanent [30], Gliding-Vertex [31], Heat-RCNN [32], CBNet [33], SESNet [17] and Kfiou [34] are introduced for comparison. Among the compared vehicle detection methods, the Gliding-Vertex, Heat-RCNN and CBNet achieved similar good AP and mAP of approximately 90%.…”

Arbitrary‐oriented traffic participant detection and axis prediction for complex crossing road in aerial view

“…Cao et al. [31] combine semantic edge detection with arbitrary‐oriented object detection to obtain more regression clues. Zheng et al.…”

“…Instead of the traditional representation, MSCNet [30] uses Gaussian distribution to alleviate the impact of the defects of the traditional oriented bounding box representation. Cao et al [31] combine semantic edge detection with arbitraryoriented object detection to obtain more regression clues. Zheng et al [28] propose a single-stage transformer detector for arbitrary-oriented detection in satellite optical imagery.…”