Colin P. Schwegmann

Very deep learning for ship discrimination in Synthetic Aperture Radar imagery

Efficient and effective ship discrimination across multiple Synthetic Aperture Radar sensors is becoming more important as access to SAR data becomes more widespread. A flexible means of separating ships from sea is ideal and can be accomplished using machine learning. Newer, advanced deep learning techniques offer a unique solution but traditionally require a large dataset to train effectively. Highway Networks allow for very deep networks that can be trained using the smaller datasets typical in SAR-based ship detection. A flexible network configuration is possible within Highway Networks due to an adaptive gating mechanism which prevents gradient decay across many layers. This paper presents a very deep High Network configuration as a ship discrimination stage for SAR ship detection. It also presents a three-class SAR dataset that allows for more meaningful analysis of ship discrimination performances. The proposed method was tested on a this SAR dataset and had the highest mean accuracy of all methods tested at 96.67%. The proposed ship discrimination method also provides improved false positive classification compared to the other methods tested.

Manifold Adaptation for Constant False Alarm Rate Ship Detection in South African Oceans

The detection of ships at sea is a difficult task made more so by uncooperative ships, especially when using transponder-based ship detection systems. Synthetic aperture radar (SAR) imagery provides a means of observation independent of the ships cooperation, and over the years, a vast amount of research has gone into the detection of ships using this imagery. One of the most common methods used for ship detection in SAR imagery is the cell-averaging constant false alarm rate (CA-CFAR) prescreening method. It uses a scalar threshold value to determine how bright a pixel needs to be in order to be classified as a ship, and thus inversely how many false alarms are permitted. This paper presents by a method of converting the scalar threshold into a threshold manifold. The manifold is adjusted using a simulated annealing (SA) algorithm to optimally fit to information provided by the ship distribution map, which is generated from transponder data. By carefully selecting the input solution and threshold boundaries, much of the computational inefficiencies usually associated with SA can be avoided. The proposed method was tested on six ASAR images against five other methods and had a reported detection accuracy (DA) of 85.2% with a corresponding FAR of 1.01 × 10-7.

Ship detection in South African oceans using a combination of SAR and historic LRIT data

Regular surveillance of territorial sea areas is increasingly important for coastal nations as these marine areas are a valuable economic resource (e.g. for fisheries or oil extraction). The responsibility for the administration, law enforcement, environmental protection and sustainable management of this frontier can be very challenging as systematic surveillance of these areas is very costly and logistically cannot cover all areas all of the time. SAR data is very popular for ship detection as large areas can be observed within a single overpass. In this paper it is shown how ship detection using the classic CFAR algorithm can be improved by using historic LRIT data.

Synthetic aperture radar ship detection using Haar-like features

The detection of ships at sea is a complex task made more so by adverse weather conditions, lack of night visibility, and large areas of concern. Synthetic aperture radar (SAR) imagery with large swaths can provide the needed coverage at a reduced resolution. The development of ship detection methods that can effectively detect ships despite the reduced image resolution is an important area of research. A novel ship detection method is introduced that makes use of a standard constant false alarm rate (FAR) prescreening step followed by a cascade classifier ship discriminator. Ships are identified using Haar-like features using adaptive boosting training on the classifier with an accuracy of 89.38% and FAR of

Small ships don't shine: Classification of ocean vessels from low resolution, large swath area SAR acquisitions

1.47 \times 10^{-8}

Ship detection in South African oceans using SAR, CFAR and a Haar-like feature classifier

across a large swath Sentinel-1 and RADARSAT-2 newly created SAR data set.

Simulated annealing CFAR threshold selection for South African ship detection in ASAR imagery

Monitoring ocean vessels that are not near the coast is difficult and expensive. One way of overcoming this is through the use of SAR satellite platforms. To monitor the largest possible area would require the use of course resolution SAR images which reduce even the largest ships to several pixels. This paper covers the datasets, methods and results used to arrive at a machine learning algorithm that classifies the size of a v1essel from course resolution SAR images. EW-GRDM Sentinel-1A images were used to classify ocean vessels by size.

Using an active contour method to detect bilge dumps from SAR imagery

Synthetic Aperture Radar images is a proven technology that can be used to detect ships at sea which have no active transponders (commonly referred to as dark targets). Various methods have been proposed that process SAR images to monitor these targets. In this paper, we propose a novel ship detection method for Advanced Synthetic Aperture Radar imagery that combines a Constant False Alarm Rate ship pre-screening method with a Haar-like feature cascade classifier. Experimental results indicate that this configuration provides a ship detection accuracy above 88% and half the False Alarm Rate of the traditional Constant False Alarm Rate method.

Ship detection in Sentinel-1 imagery using the H-dome transformation

Synthetic Aperture Radar images are able to detect ships that would be hidden to tradition ship tracking methods due to their transponders being turned off. Using a SAR image as input, the CFAR method can highlight these ships given a correctly chosen threshold value. Typically, the threshold value is chosen as a single floating value for all positions creating a flat threshold plane. This study introduces a novel method of creating a threshold plane which is adapted using Simulated Annealing. This non-flat threshold allows different areas of the image to have different threshold values thereby improving the overall performance of the ship detection system. It was found in our experiments that the proposed method improves upon the false alarm rate of the flat threshold plane CFAR method whilst keeping a similar level of detection accuracy.

Proper comparison among methods using a confusion matrix

An automatic approach to detect bilge dumping in synthetic aperture radar (SAR) images over Southern African oceans is proposed. The approach uses a threshold-based algorithm and a region-based active contour model (ACM) algorithm to achieve an efficient bilge dump detection tool. A threshold method was used to detect areas with a high bilge dump probability while the ACM method is used to get closed contours of potential bilge dumps. A discrimination step was implemented to detect linear features as potential bilge dump events. The automated approach was investigated using SENTINEL-1A and ENVISAT Advanced Synthetic Aperture Radar (ASAR) where the proposed approach showed promising results.