Sonar Image Subspace Projection for Mine Detection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional mine detection using sonar imagery relies on image processing and classification methods that often result in poor detection or classification accuracy due to their reliance on preset segmentation and feature extraction methods.
Innovation Solution
A method that projects sonar images of candidate objects onto a mine image subspace formed from known mine images, using a likelihood ratio based on probability densities indicating the probability of the object being a mine or not, with these probabilities being functions of the distance of the image to the mine image subspace, thereby improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image processing and classification methods are used, then the detection process is straightforward, but the detection accuracy is poor
Solution Approach 1:
The patent segments the high-dimensional image space into multiple lower-dimensional subspaces, each corresponding to a specific object class. This segmentation allows the system to focus detection efforts on relevant subspaces, improving detection accuracy while managing system complexity through organized decomposition of the detection task.
Solution Approach 2:
The patent transforms the detection problem from operating in the entire high-dimensional image space to operating in lower-dimensional subspaces. This dimensionality reduction approach maintains essential object characteristics while reducing computational complexity, thereby improving detection accuracy without proportionally increasing system complexity.
2Measurement precision
If preset segmentation and predetermined feature extraction methods are used, then the processing is simple, but the classification accuracy is poor
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing projection matrices for each object class subspace during system initialization. This allows the detection system to use these pre-computed projections during actual detection, improving classification accuracy while reducing processing time during operational use.
Solution Approach 2:
The patent changes the detection parameters from using predetermined feature extraction to using adaptive projections onto class-specific subspaces. This parameter change enables the system to capture more discriminative features for each object class, improving classification accuracy while the efficient projection mathematics keeps processing time manageable.
3Productivity
If the entire image space is used for detection, then all possible objects can be detected, but the processing efficiency is low
Solution Approach 1:
The patent segments the entire image space into multiple class-specific subspaces, where each subspace is dedicated to detecting a particular object class. This segmentation improves detection efficiency by focusing computational resources on relevant subspaces while maintaining the ability to detect multiple different object classes through the collection of all subspaces.
Solution Approach 2:
The patent creates a universal detection framework where the collection of class-specific subspaces together provides the ability to detect multiple object classes. Each individual subspace is specialized for one class, but the combined system maintains versatility across all object classes, achieving both efficiency and adaptability.
Data Source
AI summary
In one aspect, a method to identify a candidate object includes receiving an image of the candidate object and projecting the received image onto an image subspace. The image subspace is formed from images of known objects of a class. The method also includes determining whether the candidate object is in the object class based on the received image and the image subspace using a likelihood ratio. The likelihood ratio includes a first probability density indicating a probability an object is in the object class and a second probability density indicating a probability an object is not in the class. The first probability density and the second probability are each a function of a distance of the received image to the image subspace.