Appraisal of Low-Cost Pushbroom Hyper-Spectral Sensor Systems for Material Classification in Reflectance

Abstract

Near infrared (NIR) remote sensing has applications in vegetation analysis as well as geological investigations. For extra-terrestrial applications, this is particularly relevant to Moon, Mars and asteroid exploration, where minerals exhibiting spectral phenomenology between 600 and 800 nm have been identified. Recent progress in the availability of processors and sensors has created the possibility of development of low-cost instruments able to return useful scientific results. In this work, two Raspberry Pi camera types and a panchromatic astronomy camera were trialed within a pushbroom sensor to determine their utility in measuring and processing the spectrum in reflectance. Algorithmic classification of all 15 test materials exhibiting spectral phenomenology between 600 and 800 nm was easily performed. Calibration against a spectrometer considers the effects of the sensor, inherent image processing pipeline and compression. It was found that even the color Raspberry Pi cameras that are popular with STEM applications were able to record and distinguish between most minerals and, contrary to expectations, exploited the infra-red secondary transmissions in the Bayer filter to gain a wider spectral range. Such a camera without a Bayer filter can markedly improve spectral sensitivity but may not be necessary.

CoastalWQL: An Open-Source Tool for Drone-Based Mapping of Coastal Turbidity Using Push Broom Hyperspectral Imagery

“…Secondly, the nature of the image obtained from different sensors plays a significant role in the processing of the images. Each band image dimension from a push-broom sensor is significantly smaller than that of a snapshot imagery (Table 3 and Appendix A), as light from the observed strip of terrain is only allowed to enter through a small slit in the entrance port [24,52]. The number of images to cover a given survey area is, therefore, significantly higher due to the narrow dimension of the image and a required higher FPS (the average number of images taken by the Bayspec push-broom sensor to cover a ground coverage of around 10,000 m 2 is 30,000-40,000, compared to around 600 images with the MicaSense multispectral sensor in [21]).…”

Estimation of chlorophyll and nitrogen in double-cropping rice based on machine learning

“…Spectral analysis is widely used in the analysis of complex samples in industry, medical treatment and agriculture because of its advantages of fast analysis speed, simple operation and no need of pure samples [10][11][12][13] . Spectral data are easily interfered by stray light, noise, baseline drift and other factors, which affect the modeling effect.…”