Convolutional Network for Oil Slick Detection in Radar Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting oil slicks in radar images are time-consuming and prone to errors due to the need for human interpretation and the difficulty in distinguishing oil slicks from look-alikes, which appear frequently and resemble oil slicks in SAR images.
Innovation Solution
A computer-implemented method using a convolutional neural network trained with a combination of weighted-cross-entropy and Jaccard loss functions, incorporating leaky rectified linear units, to automatically detect oil slicks in radar images, reducing false alarms by distinguishing between oil slicks and look-alikes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human interpreters visually inspect and manually contour potential oil slick regions in SAR images, then detection accuracy can be maintained, but the process becomes time-consuming and requires experienced personnel
Solution Approach 1:
The patent replaces the mechanical human interpretation process with an automated deep learning system. A convolutional neural network (U-Net architecture) is trained to automatically segment and detect oil slicks in SAR images, eliminating the need for manual visual inspection by human interpreters while maintaining detection accuracy.
Solution Approach 2:
The system enables self-service detection through automated oil slick identification. The trained neural network independently processes SAR images to detect and contour oil slick regions without requiring experienced human interpreters, making the detection process autonomous and significantly faster.
2Reliability
If traditional detection methods are used to identify dark regions in SAR images, then oil slicks can be detected, but false alarms increase due to confusion with look-alikes such as low-wind zones, rain cells, and marine algae
Solution Approach 1:
The patent changes the detection parameters by using a trained deep learning model that learns distinguishing features of oil slicks versus look-alikes. The neural network is trained on labeled data to recognize subtle differences in texture, shape, and contextual patterns, enabling reliable differentiation between oil slicks and similar-looking features like low-wind zones or marine algae.
Solution Approach 2:
The system performs preliminary training action by pre-training the neural network on extensive labeled SAR image data containing both oil slicks and look-alikes. This preliminary learning phase enables the model to distinguish between genuine oil slicks and false alarm sources before actual detection, significantly reducing false alarm rates.
3Productivity
If supervised deep learning approaches are used to automatically detect oil slicks, then processing time is reduced, but the system requires extensive training data and computational resources
Solution Approach 1:
The patent segments the detection task into distinct computational components using a U-Net architecture with encoder-decoder structure. The encoder extracts hierarchical features through convolutional layers, while the decoder reconstructs the segmentation map. This segmentation of the neural network into modular components manages complexity while enabling fast automated detection.
Data Source
Figure 1~3
Figure 2
AI summary
The present disclosure relates to a computer implemented method (10) for detecting an oil slick in a target image acquired by a spaceborne or airborne radar, wherein said method comprises: - a phase (T1) of training a convolutional network using a set of training images, the set of training images comprising training images without oil slicks and training images with oil slicks, - a phase (T2) of predicting the presence or absence of an oil slick on the target image by applying the convolutional network on said target image, wherein the phase (T1) of training of the convolutional network uses a loss function which combines a weighted-cross-entropy loss function and a Jaccard loss function.