Data-Driven Artificial Intelligence for Calibration of Hyperspectral Big Data

Abstract

Near-earth hyperspectral big data present both huge opportunities and challenges for spurring developments in agriculture and high-throughput plant phenotyping and breeding. In this article, we present data-driven approaches to address the calibration challenges for utilizing near-earth hyperspectral data for agriculture. A data-driven, fully automated calibration workflow that includes a suite of robust algorithms for radiometric calibration, bidirectional reflectance distribution function (BRDF) correction and reflectance normalization, soil and shadow masking, and image quality assessments was developed. An empirical method that utilizes predetermined models between camera photon counts (digital numbers) and downwelling irradiance measurements for each spectral band was established to perform radiometric calibration. A kerneldriven semiempirical BRDF correction method based on the Ross Thick-Li Sparse (RTLS) model was used to normalize the data for both changes in solar elevation and sensor view angle differences attributed to pixel location within the field of view. Following rigorous radiometric and BRDF corrections, novel rule-based methods were developed to conduct automatic soil removal; and a newly proposed approach was used for image quality assessment; additionally, shadow masking and plot-level

Bioenergy relations with agriculture, forestry and other land uses: Highlighting the specific contributions of artificial intelligence and co-citation networks

“…The smart approaches may provide relevant insights [ 33 ], including on biohydrogen production with algal [ 34 ]. They are used too to promote a more circular economy [ 35 ], to predict and simulate alternative scenarios [ 36 ], as well as to analyse data [ 37 ] with useful outputs for agricultural and bioenergy productions [ 38 ]. Considering the vulnerability of the sources of biomass to the soil, weather conditions [ 39 ] and farming management [ 40 ], the prediction models play a crucial role in these frameworks.…”

Advances in field-based high-throughput photosynthetic phenotyping

“…The drone- and airborne-based sensing platforms can help relieve the time constraints ( Camino et al , 2019 ; Suarez et al , 2021 ) but may have payload limitations that need to be resolved. Additionally, as canopy-level measurements are scaled up, a large volume of data can be expected ( Sagan et al , 2021 ) and pose difficulties to manage and process ( Fu et al , 2020 ; Meacham-Hensold et al , 2020 ). Large differences are also observed in models built using leaf-level hyperspectral reflectance and those using canopy-level hyperspectral imaging for the same field trials ( Fig.…”

What Does TERRA-REF's High Resolution, Multi Sensor Plant Sensing Public Domain Data Offer the Computer Vision Community?

“…In the case of hyperspectral images, the white sensor box and scans spread out over multiple days confounded an already challenging problem. Radiometric calibration of images taken by the two hyperspectral cameras exemplifies these challenges, and a robust solution is described by Sagan et al [20] and implemented in [18]. Even processing images from an RGB camera was challenging due to fixed settings resulting in with high variability in quality and exposure, requiring the novel approach described by Li et al [17].…”