Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V20/00—Scenes; Scene-specific elements
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
  • G06V2201/08—Detecting or categorising vehicles

Definitions

• the present invention relates to an application for Identification of marine vessels using at least one image
• Identifying a marine vessel (e.g. determining whether a marine vessel is a cargo ship a tanker, or a military ship) using an image thereof obtained by means of an image capturing device can be performed with the aid of image processing, techniques.
• This representing procedure is of high Importance particularly in the defense industry.
• typically deep learning-based applications are used for representing an object visually. According to deep learning- based applications, first the features/descriptors belonging to various objects are extracted In a learning step, and then the objects are classified using said features/descriptors.
• the features of the objeca are extracted from the image of that object and then it Is calculated on a probability basis as to which class or classes acquired In the learning step the extracted feature Information may belong to Thus, the respective object is classified.
• VGG-F descriptors for descriptor extraction purposes is reported In the papers by Chatfield. K. Simonyan, K., Vedaldi A.. Zissermam .A. , "Return of the devil in the details' Delving deep into convolutional nets.” and Vedaldi, A,. Lenc, K. "Matconvnet - convolutional neural networks for MATLAB.”
• VAIS A Dataset for Recognizing Maritime Imagery- In the Visible and Infrared Speetrums
• a visible image and an infrared image are used for recognising marine vessels.
• deep convoiuiionai neural networks and gnostic fields are used in combination, in said method, however, vessel classrfication cannot be performed sufficiently accurately
• an identification method for identifying marine vessels using, at least one image.
• This method comprises the steps cf acquiring the images of a large number of marine vessels of different classes from at least one source and saving the images to a database; generating at least two superclasses according to the warine vessels in the saved images, obtaining models of said superclasses by means of deep teaming; extracting the descriptors belonging to at least one query image of at least one marine vessel and determining as to which superclass, associated with said models, the descriptors belong to. such that the marine vessel of said query image Is classified,
• the learning process can be carried out using images of different marine vessels. Additionally, since the images are separated into superclasses, a more accurate classification of the marine vessels can be provided. Furthermore, the Identification method according to the present invention makes if possible to recognize a marine vessel of a given query image among other manne vessels belonging lo the same class and to provide the recognition information to a user, to retrieve those images which are similar to a query Image, and to determine if two different, query images belong to trie same marine vessel.
• the object of the present invention is to develop an identification method for identifying marine vessels using at least one image.
• Another object of the present invention is to develop an identification method With a high accuracy rate.
• a fuiiher object of fie present invention is to develop an identification method aiding to detect if two images do belong to the same marine vehicle.
• a different object of the present invention is to develop an Identification method allowing to retrieve similar images by using an image of a marine vehicle. Description of Drawings
• FIG. 1 is flowchart of the identification method according to the present: Invention
• Figure 2 is a comparative graph, showing the mean average precision versus the number of retrieved images of the present identification method and a conwhtlonai application.
• Determining the class of a marine vessel using at least one image of that marine vessel is quite important In terms of the defense field. For Instance, as a result of identifying whether a marine vessel is a military ship or a cargo ship, any required measures pan da taken beforehand. A. rapid and accurate identification of a marine vessel makes it possible to lake the required measures in a timely and correct manner. Accordingly with the present Invention, an identification method has been developed for identifying marine vessels using at least one Image.
• the Identification method developed according to the present invention and of which an exemplary flowchart is given in Figure 1 comprises the steps of acquiring the images of a large number of marine vessels of different classes from at least one source (e.g. the Internet) (1 ) and saving the Images to a database; generating at least two superclasses according to the marine vessels in the saved images (2); obtaining models belonging to said superclasses by means of deep learning (4); extracting ihe descriptors belonging to at least one query image (I), of at least one marine vessel and determining as to which superclass, associated with said models, said descriptors belong to, such that the marine vessel of said query image (1) Is classified (5).
• a source e.g. the Internet
• the descriptors of the marine vessels In the images collected to generate said superclass (2) are extracted and the size of descriptors is reduced by principal components analysis . Then, the classes, which have similar distributions according to an interclass distance matrix and which are also related to each other semantical are combined each into a superclass by a spectral clustering method
• the collected Images were separated Into datssets In the form of a training set and a test set (3)
• the vessels in the test dataset are classified using the models learned in said training dataset
• the identification method comprises in the set separation step (3) another step of generating training and test datssets. using representations made from models of image pairs belonging to the same marine vessels (positive) and different marine vessels (negative).
• 91 ,01% verification success can be provided using the phoetpal components analysis and a support vector machine.
• images which are similar to a given query image (I) ere retrieved from an available large dataset (8) are extracted and the images, among these present In the database, the descriptors of which are close to the descriptors of the query image (1), are provided to the user in Figure Z a comparison Is given comparing the simitar image retrieval (61 application of the method according to the present invention to a conventional application (VGG-F application)...According, to the graph given ImFignre 2 the probability of retrieving Images of the same class with that of the query image (I) is higher than with the conventional application, particularly when a tow number of simliar images are retrieved, According : to another preferred embodiment of the: present invention, a marine vessel ss recognised (7) using its query image (I).
• identification information e.g. IMO-International. Maritime Organization number
• identification information e.g. IMO-International. Maritime Organization number
• the descriptors: of said query image (I) are extracted, and first the class of the marine vessel ss determined- then It as determined which, the image(s) In the determined class belong(s) to the marine vessel . Thereafter, the Identification Information belonging to the determined Image Is provided to the user and the recognition process (7) is realized accordingly,.
• the identification method according: to the present Invention makes It possible to recognize (7) a marine vessel of a given query Image (I) among other marine vessels feeionging to the same class and to provide the recognition information to a user, to receive those images which are similar to a query Image (I), and to determine if two different query Images (1) belong to the same marine vessel

Landscapes

• Engineering & Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Multimedia (AREA)
• Theoretical Computer Science (AREA)
• Image Analysis (AREA)
• Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Applications Claiming Priority (2)

Publications (1)

Family

ID=57485848

Family Applications (1)

Country Status (2)

Cited By (8)

• 2016
  • 2016-10-07 TR TR2016/14118A patent/TR201614118A1/tr unknown
  • 2016-10-10 WO PCT/TR2016/050374 patent/WO2018067080A1/en active Application Filing

Non-Patent Citations (6)

Also Published As

Similar Documents

Legal Events