Automated optical inspection of FAST’s reflector surface using drones and computer vision

The Five-hundred-meter Aperture Spherical radio Telescope(FAST)is the world’s largest single-dish radio telescope.Its large reflecting surface achieves unprecedented sensitivity but is prone to damage,such as dents and holes,caused by naturally-occurring falling objects.Hence,the timely and accurate detection of surface defects is crucial for FAST’s stable operation.Conventional manual inspection involves human inspectors climbing up and examining the large surface visually,a time-consuming and potentially unreliable process.To accelerate the inspection process and increase its accuracy,this work makes the first step towards automating the inspection of FAST by integrating deep-learning techniques with drone technology.First,a drone flies over the surface along a predetermined route.Since surface defects significantly vary in scale and show high inter-class similarity,directly applying existing deep detectors to detect defects on the drone imagery is highly prone to missing and misidentifying defects.As a remedy,we introduce cross-fusion,a dedicated plug-in operation for deep detectors that enables the adaptive fusion of multi-level features in a point-wise selective fashion,depending on local defect patterns.Consequently,strong semantics and fine-grained details are dynamically fused at different positions to support the accurate detection of defects of various scales and types.Our AI-powered drone-based automated inspection is time-efficient,reliable,and has good accessibility,which guarantees the long-term and stable operation of FAST.

Deep learning reconstruction enables full-Stokes single compression in polarized hyperspectral imaging

Polarized hyperspectral imaging,which has been widely studied worldwide,can obtain four-dimensional data including polarization,spectral,and spatial domains.To simplify data acquisition,compressive sensing theory is utilized in each domain.The polarization information represented by the four Stokes parameters currently requires at least two compressions.This work achieves full-Stokes single compression by introducing deep learning reconstruction.The four Stokes parameters are modulated by a quarter-wave plate(QWP)and a liquid crystal tunable filter(LCTF)and then compressed into a single light intensity detected by a complementary metal oxide semiconductor(CMOS).Data processing involves model training and polarization reconstruction.The reconstruction model is trained by feeding the known Stokes parameters and their single compressions into a deep learning framework.Unknown Stokes parameters can be reconstructed from a single compression using the trained model.Benefiting from the acquisition simplicity and reconstruction efficiency,this work well facilitates the development and application of polarized hyperspectral imaging.

Super-resolution compressive spectral imaging via two-tone adaptive coding

Coded apertures with random patterns are extensively used in compressive spectral imagers to sample the incident scene in the image plane.Random samplings,however,are inadequate to capture the structural characteristics of the underlying signal due to the sparsity and structure nature of sensing matrices in spectral imagers.This paper proposes a new approach for super-resolution compressive spectral imaging via adaptive coding.In this method,coded apertures are optimally designed based on a two-tone adaptive compressive sensing(CS)framework to improve the reconstruction resolution and accuracy of the hyperspectral imager.A liquid crystal tunable filter(LCTF)is used to scan the incident scene in the spectral domain to successively select different spectral channels.The output of the LCTF is modulated by the adaptive coded aperture patterns and then projected onto a lowresolution detector array.The coded aperture patterns are implemented by a digital micromirror device(DMD)with higher resolution than that of the detector.Due to the strong correlation across the spectra,the recovered images from previous spectral channels can be used as a priori information to design the adaptive coded apertures for sensing subsequent spectral channels.In particular,the coded apertures are constructed from the a priori spectral images via a two-tone hard thresholding operation that respectively extracts the structural characteristics of bright and dark regions in the underlying scenes.Super-resolution image reconstruction within a spectral channel can be recovered from a few snapshots of low-resolution measurements.Since no additional side information of the spectral scene is needed,the proposed method does not increase the system complexity.Based on the mutual-coherence criterion,the proposed adaptive CS framework is proved theoretically to promote the sensing efficiency of the spectral images.Simulations and experiments are provided to demonstrate and assess the proposed adaptive coding method.Finally,the underlying concepts are extended to a multi-channel method to compress the hyperspectral data cube in the spatial and spectral domains simultaneously.

Super-resolution compressive spectral imaging via two-tone adaptive coding:publisher’s note

This publisher’s note corrects the authorship in Photon.Res.8,395(2020).