Abstract:Because small targets have fewer pixels and carry fewer features, most target detection algorithms cannot effectively use the edge information and semantic information of small targets in the feature map, resulting in low detection accuracy, missed detections, and false detections from time to time. To solve the shortcoming of insufficient information features of small targets in the RetinaNet, this work introduces a parallel-assisted multi-scale feature enhancement module MFEM (Multi-scale Feature Enhancement Model), which uses dilated convolution with different expansion rates to avoid multiple down sampling. MFEM avoids information loss caused by multiple down sampling, and at the same time helps to assist shallow extraction of multi-scale context information. Additionally, this work adopts a backbone network improvement plan specifically designed for target detection tasks, which can effectively save small target information in high-level feature maps. The traditional top-down pyramid structure focuses on transferring high-level semantics from the top to the bottom, and the one-way information flow is not conducive to the detection of small targets. In this work, the auxiliary MFEM branch is combined with RetinaNet to construct a model with a bidirectional feature pyramid network, which can effectively integrate the strong semantic information of the high-level network and high-resolution information regarding the low level. The bidirectional feature pyramid network designed in this work is a symmetrical structure, including a top-down branch and a bottom-up branch, performs the transfer and fusion of strong semantic information and strong resolution information. To prove the effectiveness of the algorithm FE-RetinaNet (Feature Enhancement RetinaNet), this work conducts experiments on the MS COCO. Compared with the original RetinaNet, the improved RetinaNet has achieved a 1.8% improvement in the detection accuracy (mAP) on the MS COCO, and the COCO AP is 36.2%; FE-RetinaNet has a good detection effect on small targets, with APs increased by 3.2%.

Image Processing: Facilitating Retinanet for Detecting Small Objects

Tiny-RetinaNet: A One-Stage Detector for Real-Time Object Detection

Deconv R-Cnn For Small Object Detection On Remote Sensing Images

BFE-Net: Bidirectional Multi-Scale Feature Enhancement for Small Object Detection

Small Object Detection Network Based on Feature Information Enhancement

Research on Feature Enhancement for Small Object Detection

FE-RetinaNet: Small Target Detection with Parallel Multi-Scale Feature Enhancement

Small Object Detection with Improved CenterNet

Feature Implicit Enhancement Via Super-Resolution for Small Object Detection.

An attention-based feature pyramid network for single-stage small object detection

DSFPAP-Net: Deeper and Stronger Feature Path Aggregation Pyramid Network for Object Detection in Remote Sensing Images

A novel feature enhancement module for small object detection

An Efficient Object Detection Framework with Modified Dense Connections for Small Objects Optimizations

DFE-Net: detail feature extraction network for small object detection

Small Object Detection using Multi-scale Feature Fusion and Attention

Feature Rescaling and Fusion for Tiny Object Detection

Small Object Detection Using Deep Feature Pyramid Networks

An Improved Faster R-CNN for Small Object Detection

Detail Preservation and Feature Refinement for Object Detection

Extended Feature Pyramid Network for Small Object Detection

Receptive Field Fusion RetinaNet for Object Detection