5:15 PM - 6:30 PM
[STT36-P06] Ship Detection Using Synthetic Aperture Radar
Keywords:Synthetic aperture radar, Ship Detection, Adaptive threshold method
Synthetic aperture radar image includes waves, internal waves, vessel detection, and tides. With regard to the ship detection, the synthetic aperture radar PALSAR-2 mounted on ALOS-2 and AIS (Automatic Identification System), which transmits and receives identification codes, ship names, positions, speeds, etc., will cooperate. Surveillance is being conducted to identify suspicious vessels.
In last year's study, we measured the average (σ0Avg.) of the backscatter coefficient (σ0) in a small area and determined the area with a coefficient above a threshold (σth) as a vessel. As a result, the threshold of the directional scattering coefficient was set to -8.0 dB, and the number of detected vessels with AIS (Automatic Identification System) location information was 27/29 (95%) and 25/25 (100%), and many other points that were considered to be vessels were detected. In addition, we found problems related to the determination of the threshold value and the appearance of false detection points mainly in coastal areas. In this study, we compared the two methods in order to improve the detection accuracy.
The first is the fixed threshold method. We improved the threshold calculation method compared to the algorithm used in last year's study. We cut out a 500 x 500 pixel area (1) that shows one large ship and a 1000 x 1000 pixel area (2) that shows five small ships as samples from the image, and determined the threshold value based on the histograms of the sea surface and ships in areas (1) and (2). We applied this threshold to the entire image and detected pixels with a value greater than the threshold, which we considered to be ships.
The second method is the adaptive threshold method (ATM). In this method, a small region is set in the image and the threshold is calculated within this region. By scanning the entire image with this small region: the moving window, a threshold value is calculated for each region and the vessel is detected. In this study, we created an algorithm by referring to the formula for deriving the moving window and threshold used in the study by Mizukoshi, Ouchi, and Watanabe (2019).
We used one scene of November 18, 2019 in the waters from the Pacific Ocean in Wakayama Prefecture to the Kii Suido and Osaka Bay, taking into account the number of vessels traveling and the expected speed of navigation. At the same time, AIS information of the same area was used to compare ship detection .
We determined the threshold for the fixed threshold method to be DN=5000 based on the histograms of regions (1) and (2).The results of AIS and visual detection of the vessels were 33/33 (100%). The processing time of the algorithm was about 2 seconds.We set the ATM moving window as background window 81×81 and buffer window 41×41. The detection result was 33/33 ships (100%). The processing time of the algorithm was about 3 hours. The advantages of the fixed threshold method were the small number of false positives on the sea surface and the short processing time. In contrast, the advantages of ATM were clearer detection of small vessels and fewer false positives in coastal areas.
In last year's study, we measured the average (σ0Avg.) of the backscatter coefficient (σ0) in a small area and determined the area with a coefficient above a threshold (σth) as a vessel. As a result, the threshold of the directional scattering coefficient was set to -8.0 dB, and the number of detected vessels with AIS (Automatic Identification System) location information was 27/29 (95%) and 25/25 (100%), and many other points that were considered to be vessels were detected. In addition, we found problems related to the determination of the threshold value and the appearance of false detection points mainly in coastal areas. In this study, we compared the two methods in order to improve the detection accuracy.
The first is the fixed threshold method. We improved the threshold calculation method compared to the algorithm used in last year's study. We cut out a 500 x 500 pixel area (1) that shows one large ship and a 1000 x 1000 pixel area (2) that shows five small ships as samples from the image, and determined the threshold value based on the histograms of the sea surface and ships in areas (1) and (2). We applied this threshold to the entire image and detected pixels with a value greater than the threshold, which we considered to be ships.
The second method is the adaptive threshold method (ATM). In this method, a small region is set in the image and the threshold is calculated within this region. By scanning the entire image with this small region: the moving window, a threshold value is calculated for each region and the vessel is detected. In this study, we created an algorithm by referring to the formula for deriving the moving window and threshold used in the study by Mizukoshi, Ouchi, and Watanabe (2019).
We used one scene of November 18, 2019 in the waters from the Pacific Ocean in Wakayama Prefecture to the Kii Suido and Osaka Bay, taking into account the number of vessels traveling and the expected speed of navigation. At the same time, AIS information of the same area was used to compare ship detection .
We determined the threshold for the fixed threshold method to be DN=5000 based on the histograms of regions (1) and (2).The results of AIS and visual detection of the vessels were 33/33 (100%). The processing time of the algorithm was about 2 seconds.We set the ATM moving window as background window 81×81 and buffer window 41×41. The detection result was 33/33 ships (100%). The processing time of the algorithm was about 3 hours. The advantages of the fixed threshold method were the small number of false positives on the sea surface and the short processing time. In contrast, the advantages of ATM were clearer detection of small vessels and fewer false positives in coastal areas.