Simultaneous regression‐based spatial coverage estimation and object detection with deep learning

Rasmus Eckholdt Andersen,Lazaros Nalpantidis,Ole Ravn,Evangelos Boukas

doi:10.1049/ell2.12183

Rasmus Eckholdt Andersen, Lazaros Nalpantidis + Show 2 more

Open Access

https://doi.org/10.1049/ell2.12183

Copy DOI

Journal: Electronics Letters	Publication Date: Apr 16, 2021
Citations: 5	License type: CC BY 4.0

Affiliation: Technical University of Denmark

Abstract

Object detection has been in the focus of researchers within varying applications propelled by the recent advances in deep learning and neural networks. Many applications require both detection of class instances as well as a quantification of the spatial coverage of the class instances. While the performance of deep learning approaches for these tasks has been extensively studied there has not been much effort into creating a unified network structure to achieve both goals. The purpose of this paper is to present a regressor to the faster R-CNN architecture that can help quantify the spatial coverage estimation of some detected object. The goal of the regressor is to provide a reproducible result of the spatial coverage. To demonstrate the developed architecture, an example use-case of land cover estimation is used. The experiments conducted in this paper show that the network does not sacrifice object detection accuracy, and indicate that the network is able to estimate the spatial coverage of six different types of land.

Highlights

The ability to quantify areas of interest to get an estimation of how large an area is affected by a given class instance is intrinsic to inspection processes and even occurs in some surveying applications
Though object detection is capable of analysing large quantities of image data, it does not cover the use cases in which the spatial coverage is of interest rather than the actual object
There are several use-cases where performing spatial coverage is of interest to track progress or for inspection processes

Summary

Introduction

The ability to quantify areas of interest to get an estimation of how large an area is affected by a given class instance is intrinsic to inspection processes and even occurs in some surveying applications. By merging the object detection and any quantification network into a single network, the number of parameters that have to be trained get significantly reduced, compared to having separate networks for the two tasks, since a large number of parameters can be shared Another approach to solve this problem is to perform full segmentation and count the number of pixels in the image for a class instance to obtain a coverage. A VGG16 [5] backbone was used to extract features of surfaces on marine vessels containing corrosion and trained a faster R-CNN network to output bounding boxes around corrosion Their approach is limited to only localising the corrosion, and not determining the spread. ELECTRONICS LETTERS wileyonlinelibrary.com/iet-el rpn roi and classification spatial coverage

Barren

Findings

Conclusion