Citation: Wei, Y.; Wei, W.; Zhang, Y.
EfferDeepNet: An Efficient Semantic
Segmentation Method for Outdoor
Terrain. Machines 2023, 11, 256.
https://doi.org/10.3390/
machines11020256
Academic Editor: Dan Zhang
Received: 3 January 2023
Revised: 26 January 2023
Accepted: 7 February 2023
Published: 9 February 2023
Copyright: © 2023 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Article
EfferDeepNet: An Efficient Semantic Segmentation Method for
Outdoor Terrain
Yuhai Wei , Wu Wei * and Yangbiao Zhang
College of Automation Science and Engineering, South China University of Technology,
Guangzhou 510641, China
* Correspondence: weiwu@scut.edu.cn
Abstract:
The recognition of terrain and outdoor complex environments based on vision sensors is
a key technology in practical robotics applications, and forms the basis of autonomous navigation
and motion planning. While traditional machine learning methods can be applied to outdoor terrain
recognition, their recognition accuracy is low. In order to improve the accuracy of outdoor terrain
recognition, methods based on deep learning are widely used. However, the network structure of
deep learning methods is very complex, and the number of parameters is large, which cannot meet
the actual operating requirements of of unmanned systems. Therefore, in order to solve the problems
of poor real-time performance and low accuracy of deep learning algorithms for terrain recognition,
this paper proposes the efficient EfferDeepNet network for pixel level terrain recognition in order to
realize global perception of outdoor environment. First, this method uses convolution kernels with
different sizes in the depthwise separable convolution (DSC) stage to extract more semantic feature
information. Then, an attention mechanism is introduced to weight the acquired features, focusing
on the key local feature areas. Finally, in order to avoid redundancy due to a large number of features
and parameters in the model, this method uses a ghost module to make the network more lightweight.
In addition, to solve the problem of pixel level terrain recognition having a negative effect on image
boundary segmentation, the proposed method integrates an enhanced feature extraction network.
Experimental results show that the proposed EfferDeepNet network can quickly and accurately
perform global recognition and semantic segmentation of terrain in complex environments.
Keywords:
EfferDeepNet network; terrain recognition; semantic segmentation; outdoor environment
1. Introduction
In recent years, more and more mobile robots have been used in unmanned ground
vehicles, logistics and distribution, services, and other fields [
1
]. The complexity of terrain
is the main factor that interferes with robots in the efficient completion of tasks. Especially
in outdoor environments, terrain characteristics can have great changes and uncertainties.
Therefore, it is very important to ensure the stable motion of the robot in complex outdoor
terrain [
2
]. It is necessary both to ensure the accuracy of terrain recognition and to meet the
real-time task of the robot.
Pixel-level terrain recognition is the semantic segmentation of the surrounding envi-
ronment and label classification of each pixel in the image to achieve the recognition of
environmental objects [
3
]. In the process of robot work, pixel-level annotation can help the
robot to identify specific objects, which is conducive to the perception of the environment.
Pixel-level terrain recognition is a key technology for autonomous navigation of robots, and
can be used for path planning and adaptive adjustment of the distance and speed of robots
or unmanned systems [
4
]. In complex outdoor environments, pixel-level terrain recognition
technology combined with deep learning algorithms can help robots to perceive the overall
environmental information more comprehensively.
The outdoor complex terrain has obvious geometric features as well as rich texture
features. For the key texture features, traditional terrain recognition methods mainly
Machines 2023, 11, 256. https://doi.org/10.3390/machines11020256 https://www.mdpi.com/journal/machines
