Methods and systems for automatic object detection from aerial imagery

• US 10,699,119 B2
• Filed: 12/20/2019
• Issued: 06/30/2020
• Est. Priority Date: 12/02/2016
• Status: Active Grant

• Current Assignee: GEOSAT Aerospace & Technology Inc.
• Original Assignee: GEOSAT Aerospace & Technology Inc.
• Inventors: Lo, Cheng-Fang, Chen, Zih-Siou, Ko, Chang-Rong, Wu, Chun-Yi, Cheng, Ya-Wen, Chen, Kuang-Yu, Wen, Hsiu-Hsien, Lin, Te-Che, Chang, Ting-Jung
• Primary Examiner(s): Safaipour, Bobbak
• Application Number:
  US16/724,041
• Publication Number:
  US 20200125823A1
• Time in Patent Office: 193 Days
• CPC Class Codes: G06T 2207/10032 Satellite or aerial image; Remote sensing
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06V 10/242 by image rotation, e.g. by 90 degrees
• CPC Class Codes: G06V 10/42 Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation
• CPC Class Codes: G06V 10/467 Encoded features or binary features, e.g. local binary patterns [LBP]
• CPC Class Codes: G06V 10/56 relating to colour
• CPC Class Codes: G06V 10/751 Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
• CPC Class Codes: G06V 10/774 Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06V 20/188 Vegetation

Abstract

Methods and systems for detecting objects from aerial imagery are disclosed. The method includes obtaining an image of an area, obtaining a plurality of regional aerial images from the image of the area, classifying the plurality of regional aerial images as a first class or a second class by a classifier, wherein: the first class indicates a regional aerial image contains a target object, the second class indicates a regional aerial image does not contain a target object, and the classifier is trained by first and second training data, wherein the first training data include first training images containing target objects, and the second training data include second training images containing target objects obtained by adjusting at least one of brightness, contrast, color saturation, resolution, or a rotation angle of the first training images; and recognizing a target object in a regional aerial image in the first class.

Claims

1. A system for detecting objects from aerial imagery, the system comprising:

memory for storing instructions;

at least one processor configured to execute the instructions to;

obtaining an image of an area;

obtaining a plurality of regional aerial images from the image of the area;

classifying the plurality of regional aerial images as a first class or a second class by a classifier, wherein;

the first class indicates a regional aerial image contains a target object,

the second class indicates a regional aerial image does not contain a target object, and

the classifier is trained by first and second training data, wherein the first training data include first training images containing target objects, and the second training data include second training images containing target objects obtained by adjusting at least one of brightness, contrast, color saturation, resolution, or a rotation angle of the first training images; and

recognizing a target object in a regional aerial image in the first class.

2. The system of claim 1, wherein classifying the plurality of regional aerial images comprises:

determining whether appearance of a target object in a training image is rotational symmetric; and

responsive to determination that the appearance of the target object is not rotational symmetric, obtaining the second training images by adjusting the rotation angle of the first training images.

3. The system of claim 1, wherein the rotation angle comprises an angle larger than a zero degree and less than 360 degrees.

4. The system of claim 1, wherein obtaining the plurality of regional aerial images comprises:

obtaining the plurality of regional aerial images in accordance with a plurality of locations detected on the image of the area based on a Digital Surface Model (DSM) image of the area and a DSM image of a target object;

or

obtaining the plurality of regional aerial images in accordance with a plurality of location candidates on the image of the area.

5. The system of claim 4, wherein:

when a ground sample distance (GSD) of the image of the area is less than or equal to a GSD threshold, obtaining the plurality of regional aerial images comprises obtaining the plurality of regional aerial images in accordance with the plurality of locations detected on the image of the area based on the DSM images of the area and the target object; and

when the GSD of the image of the area is greater than the GSD threshold, obtaining the plurality of regional aerial images comprises obtaining the plurality of regional images in accordance with the plurality of location candidates on the image of the area.

6. The system of claim 1, wherein the target objects in the first training images comprise two different sizes.

7. The system of claim 1, wherein one of the first training images contains two target objects.

8. The system of claim 1, wherein the classifier is trained by:

a first number of training images including the first and second pluralities of training images; and

a second number of training images containing non-target objects,

wherein the first number of training images is substantially equal to the second number of training images.

9. The system of claim 1, wherein the classifier is trained by:

a first number of training images including the first and second training images;

a second number of training images containing non-target objects in the image of the area; and

a third number of training images containing non-target objects that are not in the image of the area.

10. The system of claim 9, wherein the first, second, and third numbers of training images are substantially equal.

11. The system of claim 1, wherein the at least one processor is configured to execute the instructions to:

determine that the recognized target object has plant disease when an optimized soil-adjusted vegetation index on the recognized target object is less than a plant-disease threshold.

12. A method for detecting objects from aerial imagery, the method comprising:

obtaining an image of an area;

obtaining a plurality of regional aerial images from the image of the area;

classifying the plurality of regional aerial images as a first class or a second class by a classifier, wherein;

the first class indicates a regional aerial image contains a target object,

the second class indicates a regional aerial image does not contain a target object, and

the classifier is trained by first and second training data, wherein the first training data include first training images containing target objects, and the second training data include second training images containing target objects obtained by adjusting at least one of brightness, contrast, color saturation, resolution, or a rotation angle of the first training images; and

recognizing a target object in a regional aerial image in the first class.

13. The method of claim 12, wherein classifying the plurality of regional aerial images comprises:

determining whether appearance of a target object in a training image is rotational symmetric; and

responsive to determination that the appearance of the target object is not rotational symmetric, obtaining the second training images by adjusting the rotation angle of the first training images.

14. The method of claim 12, wherein obtaining the plurality of regional aerial images comprises:

obtaining the plurality of regional aerial images in accordance with a plurality of locations detected on the image of the area based on a Digital Surface Model (DSM) image of the area and a DSM image of a target object;

or

obtaining the plurality of regional aerial images in accordance with a plurality of location candidates on the image of the area.

15. The method of claim 14, wherein:

when a ground sample distance (GSD) of the image of the area is less than or equal to a GSD threshold, obtaining the plurality of regional aerial images comprises obtaining the plurality of regional aerial images in accordance with the plurality of locations detected on the image of the area based on the DSM images of the area and the target object; and

when the GSD of the image of the area is greater than the GSD threshold, obtaining the plurality of regional aerial images comprises obtaining the plurality of regional images in accordance with the plurality of location candidates on the image of the area.

16. The method of claim 12, wherein the target objects in the first training images comprise two different sizes.

17. The method of claim 12, wherein the classifier is trained by:

a first number of training images including the first and second pluralities of training images; and

a second number of training images containing non-target objects,

wherein the first number of training images is substantially equal to the second number of training images.

18. The method of claim 12, wherein the classifier is trained by:

a first number of training images including the first and second training images;

a second number of training images containing non-target objects in the image of the area; and

a third number of training images containing non-target objects that are not in the image of the area.

19. The method of claim 18, wherein the first, second, and third numbers of training images are substantially equal.

20. The method of claim 12, comprising:

determining that the recognized target object has plant disease when an optimized soil-adjusted vegetation index on the recognized target object is less than a plant-disease threshold.

21. A non-transitory computer-readable medium storing instructions which, when executed, cause one or more processors to perform operations for detecting objects from aerial imagery, the operations comprising:

obtaining an image of an area;

obtaining a plurality of regional aerial images from the image of the area;

classifying the plurality of regional aerial images as a first class or a second class by a classifier, wherein;

the first class indicates a regional aerial image contains a target object,

the second class indicates a regional aerial image does not contain a target object, and

the classifier is trained by first and second training data, wherein the first training data include first training images containing target objects, and the second training data include second training images containing target objects obtained by adjusting at least one of brightness, contrast, color saturation, resolution, or a rotation angle of the first training images; and

recognizing a target object in a regional aerial image in the first class.

22. The non-transitory computer-readable medium of claim 21, wherein classifying the plurality of regional aerial images comprises:

determining whether appearance of a target object in a training image is rotational symmetric; and

responsive to determination that the appearance of the target object is not rotational symmetric, obtaining the second training images by adjusting the rotation angle of the first training images.

23. The non-transitory computer-readable medium of claim 21, wherein the operations comprise:

determining that the recognized target object has plant disease when an optimized soil-adjusted vegetation index on the recognized target object is less than a plant-disease threshold.