Multi-scale detector optimized for small target

Abstract

The effectiveness of deep learning networks in detecting small objects is limited, thereby posing challenges in addressing practical object detection tasks. In this research, we propose a small object detection model that operates at multiple scales. The model incorporates a multi-level bidirectional pyramid structure, which integrates deep and shallow networks to simultaneously preserve intricate local details and augment global features. Moreover, a dedicated multi-scale detection head is integrated into the model, specifically designed to capture crucial information pertaining to small objects. Through comprehensive experimentation, we have achieved promising results, wherein our proposed model exhibits a mean average precision (mAP) that surpasses that of the well-established you only look once version 7 (YOLOv7) model by 1.1%. These findings validate the improved performance of our model in both conventional and small object detection scenarios.

References

[1] LiaoY R, WangH N, LinC B, et al.. Research progress of optical remote sensing image target detection based on deep learning[J]. Journal on communications, 2022, 43(5):190-203

[2] ZhangT, LiZ, SunZ, et al.. A fully convolutional anchor-free object detector[J]. The visual computer, 2023, 39(2):569-580

[3] MohammadkarimiM, MehrabiM, ArdakaniM, et al.. Deep learning-based sphere decoding[J]. IEEE transactions on wireless communications, 2019, 18(9):4368-4378

[4] LiZ, GuoQ, SunB, et al.. Small object detection methods in complex background: an overview[J]. International journal of pattern recognition and artificial intelligence, 2023, 37(2):2350002

[5] LiR, HuJ, LiS, et al.. Blind detection of communication signals based on improved YOLO3[C], 2021, New York, IEEE: 424-429

[6] VaraP N R S, D’souzakevinB, BhargavavijayK. A downscaled faster-RCNN framework for signal detection and time-frequency localization in wideband RF systems[J]. IEEE transactions on wireless communications, 2020, 19(7):4847-4862

[7] WanY, LiaoZ, LiuJ, et al.. Small object detection leveraging density-aware scale adaptation[J]. The photogrammetric record, 2023, 38(182):160-175

[8] QinH, WuY, DongF, et al.. Dense sampling and detail enhancement network: improved small object detection based on dense sampling and detail enhancement[J]. IET computer vision, 2022, 16(4):307-316

[9] XiaoZ H, DongE Z, TongJ G, et al.. Light weight object detector based on composite attention residual network and boundary location loss[J]. Neurocomputing, 2022, 494: 132-147

[10] ZhangS F, WangQ, ZhuT, et al.. Detection and classification of small traffic signs based on cascade network[J]. Chinese journal of electronics, 2021, 30(4):727-735

[11] ChenS, LiZ, TangZ. Relation R-CNN: a graph based relation-aware network for object detection[J]. IEEE signal processing letters, 2020, 27: 1680-1684

[12] XuD, GuanJ, FengP, et al.. Association loss for visual object detection[J]. IEEE signal processing letters, 2020, 27: 1435-1439

[13] RedmonJ, DivvalaS, GirshickR, et al.. You only look once: unified, real-time object detection[C], 2016, New York, IEEE: 779-788

[14] RedmonJ, FarhadiA. YOLO9000: better, faster, stronger[C], 2017, New York, IEEE: 7263-7271

[15] REDMON J, FARHADI A. YOLOV3: an incremental improvement[EB/OL]. (2018-04-08) [2023-09-05]. https://arxiv.org/abs/1804.02767.

[16] BOCHKOVSKIY A, WANG C Y, LIAO H Y M. YOLOV4: optimal speed and accuracy of object detection[EB/OL]. (2020-04-23) [2023-09-05]. https://arxiv.org/abs/2004.10934.

[17] JIN H, SONG Q, HU X. Auto-Keras: efficient neural architecture search with network morphism[EB/OL]. (2018-06-27) [2023-09-05]. https://arxiv.org/abs/1806.10282v2.

[18] TanM, PangR, LeQ V. EfficientDet: scalable and efficient object detection[C], 2020, New York, IEEE: 10778-10787

[19] LiF, GaoD, YangY, et al.. Small target deep convolution recognition algorithm based on improved YOLOv4[J]. International journal of machine learning and cybernetics, 2023, 14(2):387-394

[20] BosquetB, CoresD, SeidenariL, et al.. A full data augmentation pipeline for small object detection based on generative adversarial networks[J]. Pattern recognition: the journal of the pattern recognition society, 2023, 133: 108998-109010

[21] YangZ, YuH, FengM, et al.. Small object augmentation of urban scenes for real-time semantic segmentation[J]. IEEE transactions on image processing, 2020, 29: 5175-5190

[22] LeeG, HongS, ChoD. Self-supervised feature enhancement networks for small object detection in noisy images[J]. IEEE signal processing letters, 2021, 28: 1026-1030

[23] ZhangH, DuQ, QiQ, et al.. A recursive attention-enhanced bidirectional feature pyramid network for small object detection[J]. Multimedia tools and applications, 2023, 82(9):13999-14018