HYPERSPECTRAL BASED TRAINING METHOD FOR ARTIFICIAL INTELLIGENCE BASED REMOTE SENSING DATA ANALYSES

Abstract

A method of training Artificial Intelligence (AI) algorithms for remote sensing image analyses by hyperspectral earth land surface property data analyses comprises subjecting a selected earth land property to unsupervised measurements of said property by using a hyperspectral image obtained from an area X of the earth land surface at a time Y by a remote sensing hyperspectral radiance measurement instrument, converting the obtained data into at least one set of data of said selected property using the appropriate unsupervised retrieval algorithm, using the at least one property data set obtained as training data for training the AI algorithm for multispectral images or microwave images of the area X; and applying the trained AI algorithm for the property to multispectral or microwave remote sensing data sets of a geographical area at least equal to the area X and for a time span at least equal to the time span Y.

Claims

1. A method of training Artificial Intelligence (AI) algorithms for remote sensing image analyses by hyperspectral earth land surface property data analyses, wherein the method comprises (a) selecting an earth land surface property to be analyzed; (b) subjecting the selected earth land surface property to unsupervised measurements of said property by using a hyperspectral image obtained from a defined area X of the earth land surface at a defined time Y by a remote sensing hyperspectral radiance measurement instrument; (c) converting obtained hyperspectral radiance measurement data into at least one set of data of the selected property using an appropriate unsupervised retrieval algorithm using physical based retrieval methods, wherein each property data set i is further assigned a geographic location X.sub.i and time of observation Y.sub.i. (d) using the at least one property data set obtained in (c) as training data for training the AI algorithm for multispectral images or microwave images of the defined area X; and (e) applying the thus trained AI algorithm for the property to multispectral or microwave remote sensing data sets of a geographical area equal to, or larger than, the defined area X and for a time span equal to, or larger than, the defined time span Y.

2. The method of claim 1, wherein said earth land surface property comprises one or more of specific leaf area, leaf nitrogen content, nitrogen uptake, fire disturbances, soil organic carbon content, plant/leaf chlorophyll content, plant diseases, chlorophyll content in water bodies, yellow substances in water bodies, mineral abundance in soils, humous content in soil, man-made materials, earth surface artificial materials.

3. The method of claim 1, wherein physically based unsupervised analyses of hyperspectral data are used as the earth land surface property data.

4. The method of claim 1, wherein data which are quantifiable by using hyperspectral images but show a signal in multispectral or microwave image data that is too weak for the property to be quantitatively determined by using classical (non-AI) data retrieval approaches are used as the earth land surface property data.

5. The method of claim 1, wherein data from operational multispectral or microwave remote sensing instruments with a needed geographic representativity and required temporal resolution are used as the data to which the trained AI algorithms are applied.

6. A method of improving a performance of a remote sensing instrument in retrieving earth land surface property data, wherein the method comprises using hyperspectral earth land surface property data for training Artificial Intelligence (AI) algorithms to be applied to data obtained from the remote sensing instrument.

7. The method of claim 6, wherein said hyperspectral earth land surface property comprises one or more of specific leaf area, leaf nitrogen content, nitrogen uptake, fire disturbances, soil organic carbon content, plant/leaf chlorophyll content, plant diseases, chlorophyll content in water bodies, yellow substances in water bodies, mineral abundance in soils, humous content in soil, man-made materials, earth surface artificial materials.

8. The method of claim 6, wherein the hyperspectral earth land surface property data used are physically and statistically based analyses of hyperspectral data.

9. The method of claim 6, wherein the hyperspectral earth land surface property data used comprise data which are quantifiable by using hyperspectral images but show a signal in multispectral or microwave image data which is too weak for the property to be quantitatively determined by using classical (non-AI) data retrieval approaches.

10. The method of claim 6, wherein the data to which the trained AI algorithms are applied comprise data from operational multispectral or microwave remote sensing instruments with a needed geographic representativity and a required temporal resolution.

Description

DETAILED DESCRIPTION OF EXEMPLARY EMBODIMENTS

[0075] The following examples are not presented with the aim of restricting the invention, but are presented for enabling a better understanding of the invention along the exemplifying description.

Example 1

[0076] A hyperspectral airborne image acquisition over an urban area is used to classify the roof material based on specific absorption or scattering features of this roof material. With such a classification, e.g. tiles, metal, bitumen, synthetic, gravel, concrete, and vegetation roofs can be classified with high accuracy [4]. These classified and material-labelled roof data are used as input data for training a Deep Learning neuronal network using multispectral or microwave data from an operational satellite mission (e. g. Sentinel-2) that covers the whole large urban area and not just the hyperspectral flight strip. After this training, the neuronal network can be applied to the whole urban area and the multispectral and microwave data and roof materials are provided as large scale maps.

Example 2

[0077] Another example is the use of a New Space hyperspectral sensor with dedicated mapping capabilities, e. g. for mapping roof materials, dry matter of crops or organic material in soils. Sensors of this type are relatively small and create much lower costs than classically concepted Earth Observation missions like CHIME [5]. However, they cover a limited spatial area, only, with a low temporal frequency. However, they could be used for defining the training data sets for the operational multispectral or microwave missions.

[0078] The invention was described above as such generally and, in addition, by referring to preferred embodiments and further preferred examples. The invention, however, is not restricted by and to the preferred embodiments and examples, but is defined broadly by the claims which follow.

CITATIONS

[0079] [1] Zhu, X. X.; Tuia, D.; Mou, L.; Xia, G.-S.; Zhang, L.; Xu, F.; Fraundorfer, F. Deep learning in remote sensing: A comprehensive review and list of resources. IEEE Geoscience and Remote Sensing Magazine 2017, 5, 8-36, doi:10.1109/MGRS.2017.2762307. [0080] [2] Qiangqiang Yuan, Huanfeng Shen, Tongwen Li, Zhiwei Li, Shuwen Li, Yun Jiang, Hongzhang Xu, Weiwei Tan, Qianqian Yang, Jiwen Wang, Jianhao Gao, Liangpei Zhang,
Deep learning in environmental remote sensing: Achievements and challenges, Remote Sensing of Environment, Volume 241, 2020, 111716, ISSN 0034-4257, doi.org/10.1016/j.rse.2020.111716. [0081] [3] Graf, L., Bach, H., Tiede, D. (2020): Semantic Segmentation of Sentinel-2 Imagery for Mapping Irrigation Center Pivots, Remote Sensing, 12(32):3937. doi.org/10.3390/rs12233937 [0082] [4] Heldens, W., 2019, Use of airborne hyperspectral data and height information to support urbanmicro climate characterisation, Dissertation, elib.dlr.de/64645/1/phd_heldens_ONLINE.pdf [0083] [5] Copernicus Hyperspectral Imaging Mission for the Environment—Mission Requirements Document,” ESA/ESTEC, 23 Jul. 2019, URL: http://esamultimedia.esa.int/docs/EarthObservation/Copernicus CHIME MRD v2.1 Is sued20190723.pdf

[0084] The entire disclosures of the above documents are expressly incorporated by reference herein.