Object Detection in Remote Sensing Images via Multi-Feature Pyramid Network with Receptive Field Block

Abstract

Object detection in optical remote sensing images (ORSIs) remains a difficult task because ORSIs always have some specific characteristics such as scale-differences between classes, numerous instances in one image and complex background texture. To address these problems, we propose a new Multi-Feature Pyramid Network (MFPNet) with Receptive Field Block (RFB) that integrates both local and global features to detect scattered objects and targets with scale-differences in ORSIs. We build a Multi-Feature Pyramid Module (M-FPM) with two cascaded convolution pyramids as the main structure of MFPNet, which handles object detection of different scales very well. RFB is designed to construct local context information, which makes the network more suitable for the objects detection around complex background. Asymmetric convolution kernel is introduced to RFB to improve the ability of feature attraction by adding nonlinear transformation. Then, a two-step detection network is constructed to combine the M-FPM and RFB to obtain more accurate results. Through a comprehensive evaluation of the experimental results on two publicly available remote sensing datasets Levir and DIOR, we demonstrate that our method outperforms state-of-the-art networks for about 1.3% mAP in Levir dataset and 4.1% mAP in DIOR dataset. Experimental results prove the effectiveness of our method in ORSIs of complex environments.

A Survey of Small Object Detection Based on Deep Learning in Aerial Images

“…This fusion process aims to enhance the detection performance of small targets in complex backgrounds. By integrating diverse features at different scales, the method improves the accuracy and robustness of target detection in challenging scenarios.Furthermore, the Multi-Feature Pyramid Network [68] (MFPNet) has been proposed to construct local context information using receiver field blocks (RFBs), which makes the network more suitable for target detection in complex backgrounds. In order to enhance the feature characterization capability and introduce nonlinear transformations, the proposed method incorporates an asymmetric convolution kernel within the RFB.…”

Research on object detection for small objects in agriculture: taking red bayberry as an example

“…At the beginning of the sample collection work, most of the bayberry fruits had already entered young fruit stage 2, the number of young fruit stage 1 samples was relatively small. We used geometric amplification method to perform a clockwise tilt of 10 ° and a counterclockwise tilt of 10 ° transformation on each image of young fruit stage 1 fruit [13][14]. At this time, the image was expanded to 201 to avoid weight deviation caused by uneven quantity.…”

A multi‐scale feature fusion convolutional neural network for facial expression recognition

“…Convolution with different dilation rates provides different receptive fields to obtain multi-scale feature information, and the feature information obtained by each dilated convolution is finally fused to generate the output. Receptive field block (RFB) (Yuan et al, 2021) utilized the Inception idea to increase the receptive field by incorporating the parallel multi-scale convolution with dilated convolution. Pyramid Pooling Module (PPM) was derived from the Pyramid Scene Parsing Network (PSP Net) (Chen et al, 2022), and the features extracted by parallel convolution in PPM were globally pooled and fused with the input features to enhance the ability to obtain global contextual information.…”