Systems and methods for physical detection using radio frequency noise floor signals and deep learning techniques
Systems and methods of using a machine learning model to detect physical characteristics of an environment based on radio signal data include at a radio signal receiver, collecting noise floor signal data comprising radio signal data from an environment within a predetermined proximity of the radio signal receiver; implementing a trained deep machine learning classifier that is trained to classify one or more physical characteristics of the environment based on the radio signal data; generating machine learning input based on the radio signal data collected by the radio signal receiver; receiving the machine learning input at the trained deep machine learning classifier; and generating by the trained deep machine learning model one or more classification labels identifying the one or more physical characteristics of the environment based on the noise floor signal data.
1. A system for deploying a deep machine learning classifier that classifies pre-trained sensory conditions within a physical environment based on radio frequency noise floor signal data, the system comprising:
one or more radio signal receivers that:
are set to receive radio signals from an unused radio frequency band, wherein the unused radio frequency band relates to a radio frequency band that is not used by one or more active radio signal transmitters within a predetermined distance of the one or more radio signal receivers;
collect radio signal data within the unused radio frequency band from one or more regions within the predetermined distance of each of the one or more radio signal receivers;
a machine learning-based system that includes a trained deep machine learning classifier, wherein the trained deep machine learning classifier identifies features of one or more sensory conditions and/or sensory events based on the collected radio signal data, wherein the machine learning-based system:
identifies a plurality of distinct sensory-based conditions and/or sensory-based events based on the radio signal data collected by each of the one or more radio signal receivers;
generate a radio frequency (RF) wavelength dataset for each of the plurality of distinct sensory-based conditions and/or sensory-based events, wherein the RF wavelength dataset for each of the plurality of distinct sensory-based conditions and/or sensory-based events includes radio wavelength data generated by the respective distinct sensory-based condition and/or sensory-based event;
generates a distinct classification label for the RF wavelength dataset for each of the plurality of distinct sensory-based conditions and/or sensory-based events, wherein the distinct classification label identifies the respective sensory-based condition and/or sensory-based event; and
augments the distinct classification label to the RF wavelength dataset for each of the plurality of distinct sensory-based conditions and/or sensory-based events;
during a training phase of one or more machine learning models, receives machine learning training input for training the one or more machine learning models, the machine learning training input comprising the RF wavelength dataset for each of the plurality of distinct sensory-based conditions and/or sensory-based events augmented with the distinct classification label;
after the training phase, deploys a radio signal receiver employing the one or more machine learning models that automatically detects and/or classifies events or conditions in a physical environment as one or more of the plurality of distinct sensory-based conditions and/or sensory-based events.
2. The system according to claim 1 , wherein
the one or more machine learning models output one or more classification labels based on radio signal data in the physical environment,
the one or more classification labels are used as input by one or more autonomous devices, and
in response to the one or more classification labels, triggering one or more automatic actions by the one or more autonomous devices, wherein the one or more automatic actions by the one or more autonomous devices includes one or more of recording details of the one or more pre-trained sensory conditions associated with the one or more classification labels, performing an actuation in response to detecting the one or more pre-trained sensory conditions, and generating and transmitting a communication comprising details of the one or more pre-trained sensory conditions.
3. The system according to claim 1 , further comprising:
a plurality of radio frequency receivers, wherein each of the plurality of radio frequency receivers collects radio signal data within a distinct unused radio frequency band from the one or more regions within the predetermined distance of each of the plurality of radio frequency receivers.
4. The system according to claim 1 , wherein:
the radio signals originate from one or more physical event sources that generate radio frequency signals that affect noise floor values in the one or more regions, wherein the one or more physical event sources include non-bio-logical or biological physical objects that affect noise floor values in the one or more regions monitored by the one or more radio signal receivers.
5. A method of using a machine learning model to detect physical characteristics of an environment based on radio signal data, the method comprising:
at a radio signal receiver, collecting noise floor signal data comprising radio signal data from an environment within a predetermined proximity of the radio signal receiver;
implementing a deep machine learning classifier that identifies features of one or more sensory conditions and/or sensory events is trained to classify one or more physical characteristics of the environment based on the radio signal data;
identifying a plurality of distinct sensory-based conditions and/or sensory-based events based on the radio signal data collected by each of the one or more radio signal receivers;
generating a radio frequency (RF) wavelength dataset for each of the plurality of distinct sensory-based conditions and/or sensory-based events, wherein the RF wavelength dataset for each of the plurality of distinct sensory-based conditions and/or sensory-based events includes radio wavelength data generated by the respective distinct sensory-based condition and/or sensory-based event;
generating a distinct classification label for the RF wavelength dataset for each of the plurality of distinct sensory-based conditions and/or sensory-based events, wherein the distinct classification label identifies the respective sensory-based condition and/or sensory-based event;
augmenting the distinct classification label to the RF wavelength dataset for each of the plurality of distinct sensory-based conditions and/or sensory-based events;
during a training phase of one or more machine learning models, receiving machine learning training input for training the one or more machine learning models, the machine learning training input comprising the RF wavelength dataset for each of the plurality of distinct sensory-based conditions and/or sensory-based events augmented with the distinct classification label; and
after the training phase, deploying a radio signal receiver employing the one or more machine learning models that automatically detects and/or classifies events or conditions in a physical environment as one or more of the plurality of distinct sensory-based conditions and/or sensory-based events.
6. The method according to claim 5 , wherein
the plurality of distinct sensory conditions and/or sensory events within the environment include one or more of an operation of machinery, a presence of a vehicle, a presence of a human, a time of day, and a weather condition.
7. The method according to claim 5 , further comprising:
training the deep machine learning classifier, wherein the training includes:
collecting by a radio signal receiver at a set radio frequency band one or more samples of noise floor signal data from a physical environment for at least one active sensory condition; and
providing training data to the deep machine learning classifier, wherein the training data comprises one or more portions of the one or more samples of noise floor signal data for the at least one active sensory condition.
8. The method according to claim 5 , further comprising:
preprocessing the samples of radio signal data by applying one or more signal processing functions to the collected radio signal data including one or more of:
performing fast Fourier Transform of one or more portions of the radio signal data; and
implementing a radio frequency spectrum analyzer that generates one or more graphical representations of the radio signal data,
wherein the training data comprises the one or more portions of the radio signal data that is transformed by fast Fourier Transform and the one or more graphical representations of the radio signal data.
9. The method according to claim 7 , further comprising:
generating a noise floor model for each of a plurality of distinct sensory conditions and/or sensory events based on the samples of radio signal data;
providing training data comprising the noise floor model for each of the plurality of distinct sensory conditions and/or sensory events.
10. The method according to claim 5 , wherein
the one or more machine learning models output one or more classification labels based on radio signal data in the physical environment,
the one or more classification labels are used as input by one or more autonomous devices, and
in response to collecting the one or more classification labels, triggering one or more automatic actions by the one or more autonomous devices, wherein the one or more automatic actions by the one or more autonomous devices includes one or more of recording details of the one or more sensory conditions and/or sensory events associated with the one or more classification labels, performing an actuation in response to detecting the one or more sensory conditions and/or sensory events, and generating and transmitting a communication comprising details of the one or more pre-trained sensory conditions.
11. The method according to claim 5 , further comprising:
implementing a plurality of networked radio signal receivers, wherein each of the plurality of networked radio signal receivers implements a distinctly trained deep machine learning classifier, wherein the distinctly trained deep machine learning classifier at each of the plurality of networked radio signal receivers classify different sets of pre-trained sensory conditions.
12. The method according to claim 7 , wherein
if the one or more samples of noise floor signal data from the physical environment that are collected by the radio signal receiver are insufficient for training the deep machine learning, generating additional machine learning input by modifying one or more features of the one or more samples of noise floor signal data to generate one or more samples of artificial noise signal data;
providing to the trained deep machine learning classifier training data comprising the one or more samples of artificial noise signal data.
13. The method according to claim 5 , further comprising:
at a second radio signal receiver:
collecting noise floor signal data from the environment using a different unused radio frequency band;
augmenting the machine learning input with the noise floor signal data collected by the second radio frequency receiver;
generating by the deep machine learning model one or more classification labels identifying the one or more physical characteristics of the environment based on the noise floor signal data from the radio signal receiver and the second radio signal receiver.
14. The system according to claim 1 , wherein
the system further:
identifies a plurality of proximate radio frequency (RF) bandwidths within the predetermined distance of each of the one or more radio signal receivers;
collects, using the one or more radio signal receivers, radio signal data from each of the plurality of proximate RF bandwidths;
generates a noise floor model based on the radio signal data from each of the plurality of proximate RF bandwidths; and
generates a composite RF signal model for a distinct sensory condition and/or distinct sensory event, the composite RF signal model combining the noise floor model for each of the plurality of proximate RF bandwidths,
wherein during the training phase of the one or more machine learning models, providing the composite RF signal model as machine learning input for training the one or more machine learning models to detect and/or classify the distinct sensory condition and/or distinct sensory event.