Tree Detection in Aerial Imagery Using Pixel Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for tree detection in aerial imagery suffer from high false positives and limited accuracy due to the inability to effectively differentiate trees from other natural and man-made objects, especially in geographical information systems where tree detection is crucial for urban planning and 3D modeling.
Innovation Solution
A pixel-level classification method using RGB channels, combined with texture and entropy features, and refined by graph cuts optimization, along with template matching for tree crown localization, to achieve accurate tree detection in aerial imagery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If object inference is learned in the context of multiple objects using color histogram and bag of SIFT features, then the method can handle various objects in aerial imagery, but the discriminating power is lowered and false positives increase to about 20%
Solution Approach 1:
The patent segments the tree detection problem into two distinct levels: pixel-level classification to identify tree regions, and object-level inference to locate individual tree crowns. This segmentation allows each level to focus on its specific task with dedicated features, improving overall accuracy while maintaining versatility.
Solution Approach 2:
The patent introduces object-specific features including tree templates and contextual constraints that are locally adapted to tree characteristics. These local quality enhancements allow the system to differentiate trees from other objects more effectively without sacrificing the ability to handle multiple object types.
2Measurement precision
If remote sensing technologies (LiDAR and NIR) are used for tree detection, then geometric and radiometric measures provide strong heuristic for tree detection, but the availability of such imagery is very limited compared to aerial imagery
Solution Approach 1:
The patent replaces the need for specialized remote sensing equipment (LiDAR and NIR sensors) with standard aerial imagery that is more widely available. By substituting the detection mechanism to work with readily accessible RGB aerial images, the system maintains practical applicability while achieving competitive detection accuracy through sophisticated image processing.
3Reliability
If contextual constraints are applied to resolve ambiguities in learned results, then some false positives can be reduced, but the final results still contain about 20% false positives
Solution Approach 1:
The patent performs preliminary pixel-level classification before object-level inference, pre-segmenting tree regions to reduce the search space for subsequent tree crown detection. This preliminary action simplifies the overall processing while improving reliability by eliminating obvious non-tree regions early in the pipeline.
Solution Approach 2:
The patent adds a spatial dimension to the detection process by introducing tree templates and spatial contextual constraints that operate at the object level, complementing the pixel-level classification. This dimensional enhancement provides additional discrimination power to reduce false positives without excessively increasing complexity.
Data Source
AI summary
Methods and systems for detecting trees in aerial imagery are provided. A method for detecting trees in aerial imagery may include training a pixel-level classifier to assign a tree or non-tree label to each pixel in an aerial image. The method may further include segmenting tree and non-tree regions of the labeled pixels. The method may also include locating individual tree crowns in the segmented tree regions using one or more tree templates. A system for detecting trees in aerial imagery may include a trainer, a segmenter and a tree locator.