Aerial Imagery Boundary Detection Using Discrete Wavelet Transform
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Detecting and aligning land property boundaries on aerial imagery is challenging due to thin objects like fences and walls being difficult to perceive, and existing methods suffer from irreversible information loss, especially for small-scale entities.
Innovation Solution
The system employs a computer vision architecture that uses discrete wavelet transforms to preserve high-frequency details, applying a feature encoder with convolution blocks and discrete wavelet transform layers, and a feature decoder with inverse discrete wavelet transform layers, to detect and align land property boundaries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional aerial imagery processing methods are used, then processing speed is maintained, but boundary detection precision deteriorates due to irreversible information loss
Solution Approach 1:
The patent segments the aerial imagery into multiple frequency bands using discrete wavelet transform, separating high-frequency components (which contain boundary information) from low-frequency components. This segmentation allows the system to preserve and process boundary-related information separately, preventing information loss during compression and processing operations.
Solution Approach 2:
The patent transforms the image processing from spatial domain to frequency domain using wavelet transform. By operating in the frequency dimension rather than just the spatial dimension, the system can selectively preserve high-frequency boundary information while compressing low-frequency background information, thereby maintaining detection precision while reducing overall data volume.
2Measurement precision
If high-resolution aerial imagery is used, then boundary detection precision is improved, but processing complexity increases
Solution Approach 1:
The patent divides the high-resolution image processing task into separate frequency band processing tasks. By segmenting the processing workload across different frequency bands, the system can optimize processing for each band independently, reducing overall computational complexity while maintaining the ability to detect fine boundary details.
Solution Approach 2:
The patent extracts and caches high-frequency components separately from the main processing pipeline. By taking out the boundary-critical high-frequency information and processing it independently with specialized operators, the system achieves high detection precision without requiring the full computational resources needed for processing the entire high-resolution image at maximum detail.
3Measurement precision
If thin boundary objects are detected, then alignment accuracy is improved, but detection difficulty increases
Solution Approach 1:
The patent detects thin boundary objects by transforming the detection problem from spatial domain to frequency domain. In the frequency domain, thin boundaries manifest as distinct high-frequency patterns that are easier to identify and distinguish from background noise, thereby reducing detection difficulty while improving alignment accuracy.
Solution Approach 2:
The patent applies different processing operators tailored to specific frequency bands and boundary characteristics. By adapting the detection methodology to the local properties of boundary regions (using high-pass filtering and specialized convolution operators for edge detection), the system overcomes the inherent difficulty of detecting thin objects while achieving high alignment precision.
Data Source
AI summary
Systems and methods for detecting and aligning land property boundaries on aerial imagery are provided. The system receives an aerial imagery having land properties. The system applies a feature encoder having a plurality of levels to the aerial imagery. A first level of the plurality of levels includes a convolution block and a discrete wavelet transform layer. The discrete wavelet transform layer decomposes an input feature tensor to the first level into a low-frequency band and a high-frequency band. The high-frequency band is cached and processed with side-convolutional blocks before the high-frequency band are passed to a feature decoder. The system applies the feature decoder to an output of the feature encoder based at least in part on one of inverse discrete wavelet transform layers. The system determines boundaries of the one or more land properties based at least in part on a boundary cross-entropy loss function.