SAR Ship Classification Using Projection-Aware Image Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ship classification algorithms using SAR images are inadequate in distinguishing various ship types due to changes in SAR geometry, leading to similar appearances of ships from different classes.
Innovation Solution
An image processing apparatus that utilizes projection calculator units to determine incident and foreshortening angles, normalizes pixel values, and employs feature learners to extract and classify objects based on learned features, incorporating SAR geometry information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional ship classification algorithms are used, then the processing speed is fast, but the classification accuracy deteriorates due to changes in SAR geometry causing similar appearances of different ship types
Solution Approach 1:
The patent transforms the classification problem from a 2D image-based approach to a 3D feature space by incorporating incident angle and foreshortening direction angle as additional dimensions. This allows the system to distinguish ships that appear similar in 2D SAR images by considering their geometric properties in three-dimensional space, thereby improving classification accuracy without excessive complexity increase
Solution Approach 2:
The patent changes the parameter space by introducing geometric parameters (incident angle, foreshortening direction angle) alongside traditional image features. By normalizing these parameters and combining them with image features through a learned mapping function, the system adapts to varying SAR geometries and improves classification accuracy across different viewing conditions
2Measurement precision
If SAR geometry information is incorporated into the classification process, then the classification accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary normalization of geometric parameters (incident angle, foreshortening direction angle) before they are fed into the classification model. This pre-processing step standardizes the input data range and reduces the computational burden during the main classification process, allowing accurate classification with reduced real-time computational power requirements
Solution Approach 2:
The patent introduces a normalization layer as an intermediary between the raw SAR image features and the classification model. This intermediary component transforms geometric parameters into a standardized format, facilitating more efficient processing by the classification algorithm and reducing overall computational complexity while maintaining accuracy
3Adaptability or versatility
If traditional feature extraction methods are used, then the processing is simple, but the ability to distinguish ships under different SAR geometries is insufficient
Solution Approach 1:
The patent implements a dynamic feature extraction approach where the mapping function learns to adaptively combine image features with geometric parameters based on the specific SAR geometry conditions. This dynamic adaptation allows the system to maintain high classification accuracy across varying incident angles and foreshortening directions, unlike static traditional methods
Solution Approach 2:
The patent creates a universal feature representation that works across different SAR geometries by combining multiple feature types (image features, incident angle, foreshortening direction angle) into a unified feature vector. This multi-functional feature set can handle various ship types and viewing conditions, providing versatility without requiring separate specialized algorithms for each scenario
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An object is to provide an image processing apparatus capable of appropriately distinguishing various object types. An image processing apparatus (1C) comprising: detector means (11) for detecting objects in an input SAR image and generating object chips; projection calculator means (12) for calculating projection information of each object using SAR geometry; feature learner means (14) for learning, for each object, a relation between an object chip and its projection information, and thereby generating learnt features of object chips; and classifier means (15) for classifying object chips into classes based on the learnt features of object chips.