Method and apparatus to classifying craft
A method of training a machine learning, ML, algorithm, is described. The method is implemented, at least in part, by a computer comprising a processor and a memory. The method comprises: providing training data comprising a set of audio signals, including a first audio signal, as respective bitstreams, corresponding with respective sets of watercraft (S 501 ); training the ML algorithm using the provided training data comprising detecting the respective sets of watercraft and classifying the detected respective sets of watercraft according to a set of classes, including a first class, based, at least in part, on a set of labels, including a first label, wherein the set of labels relates to propeller, propulsor, prime mover and/or submerged equipment associated with watercraft (S 502 ).
1 . A method of classifying watercraft, the method implemented by a computer comprising a processor and a memory, the method comprising:
providing training data comprising a set of training audio signals, including a training audio signal, as respective bitstreams, corresponding with respective sets of watercraft;
training a machine learning (ML) algorithm using the provided training data, the training comprising detecting the respective sets of watercraft and classifying the detected respective sets of watercraft according to a set of classes based on the training audio signal, including a first class, based on a set of labels, including a first label, wherein the set of labels relates to a propeller, a propulsor, a prime mover, and/or a submerged equipment associated with the respective sets of watercraft, thereby providing a trained ML algorithm configured to fit the training audio signal to the first label;
providing the computer aboard a communicatively isolated sub-surface watercraft;
obtaining a first input audio signal corresponding with a first set of watercraft, including a first watercraft;
inferring the first class of the set of classes, representing the first watercraft, using the trained ML algorithm and the first input audio signal;
displaying, on a display communicatively coupled to the computer, respective representations of the first set of watercraft, including the first watercraft, via a graphical user interface; and
displaying, on the display, the set of labels, including the first label, of the first class representing the first watercraft as a grid, each label in the set of labels being colored depending on how many of the labels associated with the first set of watercraft were fit by the trained ML algorithm to the first input audio signal.
2 . The method according to claim 1 , comprising reclassifying the first class, representing the first watercraft, responsive to re-selection of the set of labels by the human user.
3 . The method according to claim 1 , comprising reclassifying the first class, representing the first watercraft, responsive to receiving information related to the first watercraft.
4 . The method according to claim 3 , comprising training the ML algorithm using the training audio signal, based on reclassifying the first class, representing the first watercraft.
5 . The method according to claim 1 , comprising:
obtaining a second input audio signal corresponding with the first set of watercraft, including the first watercraft; and
inferring the first class of the set of classes, representing the first watercraft, using the trained ML algorithm, the obtained second input audio signal, and the obtained first input audio signal.
6 . The method according to claim 1 , comprising determining a location, a bearing, a velocity, an acceleration, a manoeuvre, and/or an activity of the first watercraft, using the trained ML algorithm and the obtained first input audio signal.
7 . The method according to claim 1 , wherein the first-training audio signal comprises frequencies in a range from 10 Hz to 30 KHz.
8 . The method according to claim 1 , wherein the first training audio signal corresponds with a set of rotating propellers, including a first rotating propeller, defined based on propeller information including at least one of a number of propellers, a type of propeller, a number of blades, a propeller characteristic, a blade characteristic, a rotational speed, or a shaft characteristic; and/or wherein the training audio signal corresponds with a set of propulsors, including a first propulsor, defined based on propulsor information including at least one of a type of propulsor or a propulsor characteristic.
9 . The method according to claim 1 , wherein the first-training audio signal corresponds with a set of prime movers, including a first prime mover, defined based on prime mover information including at least one of a type of prime mover or a prime mover characteristic.
10 . The method according to claim 1 , wherein the first-training audio signal corresponds with submerged equipment associated with the respective sets of watercraft.
11 . The method according to claim 1 , wherein providing the training data comprises obtaining the training audio signal, using a hydrophone.
12 . The method according to claim 1 , wherein providing the training data comprises generating the training audio signal, by combining two or more obtained audio signals.
13 . The method according to claim 1 , wherein providing the training data comprises creating a set of spectrograms, including a first spectrogram, from the set of training audio signals, wherein the first spectrogram corresponds with the training audio signal.
14 . A non-transient computer-readable storage medium encoded with instructions that when executed by one or more processors cause a process to be carried out for classifying watercraft, the process comprising:
training a machine learning (ML) algorithm using training data, the training data including a set of training audio signals as respective bitstreams, corresponding with respective sets of watercraft, the training comprising detecting the respective sets of watercraft and classifying the detected respective sets of watercraft according to a set of classes, including a first class, based on the training audio signals and a set of labels, including a first label, wherein the set of labels relates to a propeller, a propulsor, a prime mover, and/or a submerged equipment associated with the respective sets of watercraft, thereby providing a trained ML algorithm configured to fit the training audio signal to the first label;
obtaining a first input audio signal corresponding with a first set of watercraft, including a first watercraft;
inferring the first class of the set of classes, representing the first watercraft, using the trained ML algorithm and the first audio signal;
displaying, on a display, respective representations of the first set of watercraft, including the first watercraft, via a graphical user interface; and
displaying, on the display, the set of labels, including the first label, of the first class representing the first watercraft as a grid, each label in the set of labels being colored depending on how many of the labels associated with the first set of watercraft were fit by the trained ML algorithm to the first audio signal.
15 . The method according to claim 3 , comprising training the ML algorithm using the first training audio signal, based on reclassifying the first class, representing the first watercraft.
16 . The method according to claim 15 , comprising:
obtaining a second input audio signal corresponding with the first set of watercraft, including the first watercraft; and
inferring the first class of the set of classes, representing the first watercraft, using the trained ML algorithm, the obtained second input audio signal, and the obtained first input audio signal.
17 . The method according to claim 16 , comprising determining a location, a bearing, a velocity, an acceleration, a manoeuvre, and/or an activity of the first watercraft, using the trained ML algorithm and the obtained first input audio signal, wherein the first-training audio signal comprises frequencies in a range from 10 Hz to 30 kHz.
18 . The method according to claim 17 , wherein the first training audio signal corresponds with:
a set of rotating propellers, including a first rotating propeller, defined based on propeller information including at least one of a number of propellers, a type of propeller, a number of blades, a propeller characteristic, a blade characteristic, a rotational speed, or a shaft characteristic;
a set of propulsors, including a first propulsor, defined based on propulsor information including at least one of a type of propulsor or a propulsor characteristic;
a set of prime movers, including a first prime mover, defined based on prime mover information including at least one of a type of prime mover or a prime mover characteristic; and/or
submerged equipment associated with the respective sets of watercraft.
19 . A non-transient computer-readable storage medium encoded with instructions that when executed by one or more processors cause a process to be carried out for classifying watercraft, the process comprising:
obtaining a first input audio signal corresponding with a first set of watercraft, including a first watercraft;
obtaining a second input audio signal corresponding with the first set of watercraft, including the first watercraft; and
inferring a first class of a set of classes, representing the first watercraft, using a trained machine learning (ML) algorithm, the first input audio signal, and the second input audio signal, the trained ML algorithm trained using training data, the training data including a set of training audio signals as respective bitstreams, corresponding with respective sets of watercraft, the training comprising detecting the respective sets of watercraft and classifying the detected respective sets of watercraft according to the set of classes, including the first class, based on the training audio signals and a set of labels, including a first label, wherein the set of labels relates to a propeller, a propulsor, a prime mover, and/or a submerged equipment associated with the respective sets of watercraft, thereby providing a trained ML algorithm configured to fit the training audio signal to the first label;
displaying, on a display, respective representations of the first set of watercraft, including the first watercraft, via a graphical user interface; and
displaying, on the display, the set of labels, including the first label, of the first class representing the first watercraft as a grid, each label in the set of labels being colored depending on how many of the labels associated with the first set of watercraft were fit by the trained ML algorithm to the first audio signal.
20 . The method according to claim 1 , wherein the first audio signal is associated with a construct/contrast pair, and wherein a first label in the set of labels is colored a first color when the first label is fit to the first audio signal and a second color when the first label is not fit to the first audio signal, the first color being different from the second color.