Reducing false positive detections of gravity-based natural disasters in radar data

Abstract

Reducing false alarms of gravity-related natural disasters in radar data. The present invention relates to a computer-implemented method (100) for detecting gravity-related natural disasters, the method comprising: continuously receiving (110) radar data in real time; continuously monitoring (120) the radar data using a detection algorithm to detect (130) indications of gravity-related natural disaster events in the radar data; and, if an indication of a gravity-related natural disaster event is detected, initiating (140) an action in real time, wherein: after the indication of a gravity-related natural disaster event is detected, a trained neural network evaluates (160) the accumulated (150) radar data for at least a defined time period; the trained neural network confirms (170) or refutes the gravity-related natural disaster event based on the evaluation (160); and if the gravity-related natural disaster event is refuted, automatically revising (180) the initiated action.

Description

Technical Field

[0001] This invention generally relates to the field of natural disaster detection. More specifically, this invention relates to the field of using radar systems to detect gravity-based natural disasters. Background Technology

[0002] Gravity-related natural disasters include, but are not limited to, events such as avalanches, landslides, mudflows, debris flows, and rockfalls. These disasters occur when rocks, soil, or snowballs slide or fall down a slope due to gravity. Historically, the detection of such disasters relied on physical sensors and human observation. These methods typically involve seismic sensors, infrasound sensors, or visual monitoring. While effective to some extent, these methods can be limited by environmental conditions and require significant human intervention. Recently, radar systems have been used to detect gravity-related natural disasters by analyzing distance and movement approaching or receding within the radar's field of view. These systems operate effectively under various weather conditions, including fog and rain / snow, in which vision- or laser-based systems would cease to function.

[0003] The downside is that all these methods, including radar-based ones, struggle to distinguish between real-world events and false alarms (e.g., false alarms caused by noise, interference, or environmental factors), leading to unnecessary alerts and resource allocation. The goal is to reduce the number of false alarms. Some solutions have been implemented using basic machine learning algorithms (e.g., decision trees or support vector machines) to improve detection accuracy. However, these methods fail to capture complex patterns in radar data. Furthermore, some methods require significant human intervention, making them poorly scalable and inefficient, while others require manual feature engineering and do not adapt well to new data. Summary of the Invention

[0004] Therefore, one object of the present invention is to provide an improved computer-implemented method and an improved system for radar-based detection of gravity-based natural disasters.

[0005] Another object of the present invention is to provide a method and system that have enhanced detection accuracy, thereby reducing false alarms.

[0006] One specific objective is to provide a method and system that can reduce the number of false alarms in radar-based gravity-based natural disasters, thereby reducing resource allocation due to unnecessary alerts.

[0007] Another objective is to provide a method and system that can work without user intervention—for example, work completely autonomously.

[0008] At least one of these objectives is achieved by a computer-implemented method and system according to the present invention.

[0009] The proposed solution utilizes an artificial neural network to process accumulated radar data to determine the likelihood of actual gravity-related natural disaster events, such as avalanches or rockfalls. The network is trained to recognize patterns and distinguish between real events and false alarms. By reassessing the situation using more data, the proposed method and system minimize the impact of false alarms on resources and infrastructure, reducing unnecessary interventions. For example, this can significantly reduce unnecessary measures such as road closures, thereby improving traffic flow and reducing inconvenience to the public. The solution includes a two-stage assessment, where the combination of immediate response and delayed confirmation is a key feature, providing both rapid action and ensuring accuracy. The use of neural networks allows the system to adapt and learn from new data over time, improving long-term performance with minimal human intervention. The proposed solution can be generalized to different locations by training with more data collected from different locations over time. This enables it to accurately distinguish between real events of various scales and common false alarms, such as those caused by weather-related noise in the atmosphere.

[0010] A first aspect of the present invention relates to a computer-implemented method for detecting gravity-type natural disasters, the method comprising: continuously receiving radar data in real time; continuously monitoring the radar data using a detection algorithm to detect indications of gravity-type natural disaster events in the radar data; and, if an indication of a gravity-type natural disaster event is detected, initiating an action in real time. Furthermore, after detecting an indication of a gravity-type natural disaster event, a trained neural network evaluates radar data accumulated over at least a defined time period, and confirms or refutes the gravity-type natural disaster event based on the evaluation. If the gravity-type natural disaster event is refuted, the initiated action is automatically revised.

[0011] For example, the gravity-type natural disaster event can be one of the following: avalanche, landslide, mudslide, debris flow, and rockfall.

[0012] According to some implementations, the method further includes accumulating received radar data, for example, wherein the received radar data is stored in a data memory and retrieved from the data memory for evaluation.

[0013] In some implementations, the accumulated radar data begins upon detection of an indication of a gravity-related natural disaster event (e.g., immediately after detection). For example, the defined time period begins from the point in time when the indication of a gravity-related natural disaster event is detected.

[0014] In other implementations, the accumulated radar data includes continuously accumulated radar data. For example, the defined time period begins before the detection of a gravity-based natural disaster event indication and ends after the detection of the indication.

[0015] According to some embodiments of the method, the radar data is received from a radar system configured as a Doppler radar.

[0016] According to some implementations of the method, the detection algorithm is a rule-based algorithm.

[0017] According to some embodiments of the method, initiating the action includes sending commands to the alarm infrastructure and / or remotely controlling one or more components of the alarm infrastructure. For example, the alarm infrastructure includes: - One or more traffic lights and / or barriers, wherein activating the action includes controlling the traffic lights and / or barriers to close a road or railway track. - One or more alarms, wherein activating the action includes controlling the alarm to trigger an audible emergency signal, and / or - One or more warning light signals, wherein initiating the action includes controlling the warning lights to trigger a visual emergency signal.

[0018] According to some implementations, if the occurrence of a gravity-related natural disaster has been confirmed, the method includes automatically initiating further actions. For example, initiating these further actions includes: - Send commands to the alerting infrastructure. - One or more components of a remotely controlled alarm infrastructure, and / or - Send one or more messages to personal devices located in areas potentially affected by gravity-type natural disasters, particularly via one of GSM, Wi-Fi, fiber optic, radio, satellite communication, and / or LoRa-WAN.

[0019] According to some embodiments of the method, the trained neural network also evaluates further data from one or more other sensors. In some embodiments, the other sensors include one or more of a seismic sensor, an infrasound sensor, a camera, an infrared sensor, and a microphone. In some embodiments, the further data is received upon detection of a gravity-based natural disaster event indication. In some embodiments, the one or more other sensors are activated upon detection of a gravity-based natural disaster event indication. In some embodiments, the trained neural network also evaluates weather data.

[0020] According to some implementations, the method is executed in a computing unit having at least one first data connection for continuously receiving radar data from the radar system in real time. Specifically, the computing unit further includes at least one second data connection for sending commands to the alarm infrastructure to initiate the action and / or revise an already initiated action.

[0021] A second aspect of the invention relates to a system for detecting gravity-related natural disasters—for example, the method according to the first aspect. The system includes a computing unit having at least one first data connection for continuously receiving real-time radar data from a radar system and storing a detection algorithm configured to continuously monitor the radar data to detect indications of gravity-related natural disaster events in the radar data. According to this aspect of the invention, the computing unit stores a trained neural network configured to, upon detecting an indication of a gravity-related natural disaster event, evaluate radar data accumulated over at least a defined time period and, based on the evaluation, confirm or refute the gravity-related natural disaster event.

[0022] According to some embodiments, the system includes the radar system, and in particular, the radar system is configured as a Doppler radar.

[0023] The third aspect of the invention relates to a computer program product comprising program code stored on a machine-readable medium or embodied by electromagnetic waves containing program code segments, and having computer-executable instructions for performing the method according to the first aspect of the invention when executed, particularly on a computing unit of a system according to the second aspect of the invention. Attached Figure Description

[0024] The present invention will now be described in detail with reference to exemplary embodiments shown in the accompanying drawings, wherein: Figure 1 An exemplary embodiment of the system according to the present invention is shown; Figures 2a to 2c Three exemplary radar images are shown; and Figure 3 This is a flowchart illustrating an exemplary embodiment of a computer-implemented method according to the present invention. Detailed Implementation

[0025] exist Figure 1An exemplary embodiment of the system according to the present invention is illustrated. The system includes a computing unit 3 configured to continuously receive radar data from a radar unit 2 and evaluate the radar data in real time to determine whether the radar data indicates a currently occurring gravity-type natural disaster 1. If such a gravity-type natural disaster 1 is determined to have occurred, the computing unit 3 automatically initiates multiple actions. As shown, these actions may include triggering alarms in areas threatened by disaster 1 using alarm infrastructure 4. For example, a road or railway section located below a detected avalanche and therefore likely to be impacted by the avalanche in the near future. To prevent potential harm to the threatened area, it may be closed using installed traffic signals or barriers.

[0026] Gravity-related natural disasters 1 include avalanches and icefalls, as well as landslides (which further include rockfalls, rockfalls, debris flows, and slower slope instability). For example, avalanches occur suddenly and are difficult to predict. Alarm systems with automatic avalanche detection enable early detection of avalanches (i.e., at high altitudes) to trigger alarms, close roads or railways, or evacuate construction sites threatened by gravity-related natural disasters 1. Radar units 2 used as avalanche radars can include Doppler radars configured to detect moving snow masses across an entire mountainside. For example, using a 5 km measurement range and a 10 km coverage area, one avalanche Doppler radar can cover multiple avalanche paths and track and map detected avalanches. Similarly, Doppler radars can also be used to detect other types of gravity-related natural disasters 1, such as rockfalls or debris flows. A fundamental prerequisite for an alarm system is sufficient warning time between event detection and the arrival of the disaster at threatened infrastructure.

[0027] Data transmission between radar unit 2, computing unit 3, and alarm infrastructure 4 may include GSM (cellular network), Wi-Fi, fiber optic, radio (including directional radio), and / or LoRa-WAN. Preferably, at least two independent communication channels may be used.

[0028] Radar unit 2 includes one or more radar devices (e.g., Doppler radar devices) whose sensors are pointed at one or more areas where gravity-related natural disasters 1 may occur, such as slopes, hillsides, or rock walls.

[0029] The computing unit 3 can be of any type, such as a server computer, desktop computer, or portable device. A rule-based detection algorithm 5 runs on the computing unit 3 to evaluate continuously received radar data in real time and determine whether the radar data indicates an ongoing gravity-related natural disaster 1. Detection algorithm 5 is a "real-time" algorithm capable of making rapid decisions based on radar data within seconds. This is crucial for immediate response, such as activating traffic lights or barriers to close threatened roads or railways. If the rule-based detection algorithm 5 determines that the radar data indicates a gravity-related natural disaster 1, an alarm (and / or other action) is automatically triggered / initiated in real time.

[0030] Alarms can be triggered in various ways, including using traffic lights and barriers, sirens, radios with headsets for noisy environments (e.g., for construction sites in threatened areas), or by sending messages via SMS, email, or automated calls (cell broadcasts or priority services).

[0031] Furthermore, a trained neural network 6 is mounted on the computing unit 3. After the initial decision, i.e., once the rule-based detection algorithm 5 determines that the radar data indicates a gravity-type natural disaster 1, the system continues to accumulate radar data over a period of time. The trained neural network 6 then uses this accumulated radar data for a secondary evaluation to confirm or refute the initial decision of the "real-time" algorithm, i.e., the occurrence of the gravity-type natural disaster 1.

[0032] If the trained neural network 6 refutes the occurrence of disaster 1 (i.e., concludes that the decision made by the rule-based algorithm 5 is wrong), then the initiated actions (e.g., the triggered alarms) can be revised.

[0033] If the trained neural network 6 confirms the occurrence of disaster 1 (i.e., concludes that the decision of the rule-based algorithm 5 is correct), further actions can be initiated, such as sending another message to confirm that disaster 1 has indeed occurred.

[0034] Figure 2a , Figure 2b and Figure 2c Three exemplary visualizations of accumulated radar data are shown, each displaying activity measured by the radar at a distance (d) from radar element 2 over a time period (t). Darker pixels in the image indicate stronger activity. These images are derived from raw data from a Doppler radar system. Each image shows an example of a radar signal that triggered an alarm due to a rule-based algorithm 5 determining the presence of an indication of a gravity-related natural disaster 1.

[0035] exist Figure 2a In the example, a passing helicopter caused a dark pattern in the image. Figure 2bIn the case of the pattern, the noise is caused by weather-related noise in the atmosphere. In both cases, many rule-based algorithms trigger false alarms. Figure 2c An image is shown in which the dark patterns are caused by an avalanche (i.e., the current gravity-related natural disaster 1).

[0036] Such images, derived from raw data from Doppler radar systems, can be used to train neural networks. The use of neural networks allows the system to learn and adapt with continued use, thereby improving its performance. When trained with increasing amounts of data collected over time from different locations, the neural network learns to accurately distinguish real-world events of various scales from common false alarms, such as those caused by weather-related noise in the atmosphere.

[0037] Figure 3 The flowchart illustrates an exemplary embodiment of a computer-implemented method 100 for detecting gravity-based natural disasters according to the present invention. The method can be executed fully automatically—for example, by referring to… Figure 1 The system described.

[0038] 110. Radar data is continuously received from radar equipment in real time, with no significant delay. 120. The received radar data is continuously monitored using a rule-based "real-time" detection algorithm to check for indications of ongoing or apparent gravity-related natural disasters. This reception and monitoring continues until an event such as 130 is detected. In this case, 140. Action is initiated in real time, with no significant delay. This immediate response is a key feature of natural disaster detection, providing rapid action to prevent harm to people and movable objects in areas threatened by gravity-related natural disasters 1.

[0039] The actions initiated can be standardized actions for all detected events. Alternatively, radar data can be interpreted based on the type, size, or location of the detected events, allowing actions to be tailored specifically to the type, size, or location of gravity-related natural disasters. For example, if an event is detected, all individuals within a predetermined area can be alerted, such as through sirens and / or community broadcasts, or only individuals near areas expected to be affected by the natural disaster can be alerted.

[0040] Since false alarms may occur—for example, triggered by certain weather changes or flying objects within the radar unit's field of view—according to the present invention, a trained neural network is provided for secondary evaluation 160 to confirm 170 or refute the decision of the "real-time" algorithm.

[0041] When the rule-based algorithm detects event 130, radar data is still being received and accumulated over a period of time, accumulating to 150. This period needs to be long enough to be evaluated by the trained neural network, 160, for example, at least several seconds (e.g., 30 seconds). This period can be predefined.

[0042] In some implementations, accumulation 150 only begins when an event is detected. The neural network then evaluates the accumulated radar data 160, for example, after the time period has elapsed.

[0043] In other embodiments, the received data is accumulated 150 before event 130 is detected. Optionally, method 100 may include continuous data accumulation 150. In some embodiments, instead of starting data accumulation only after the "real-time" algorithm detects event 130, it includes continuously accumulating enough data for the neural network to review and make decisions. This allows the rule-based algorithm and the trained neural network to perform combined real-time analysis at the same point in time. This approach can improve the accuracy of real-time detection and may identify patterns or event precursors that might be missed by rule-based algorithms alone. In yet another embodiment, the neural network may also evaluate 160 data from before and after the detected event.

[0044] Therefore, the time period for accumulating radar data begins before the detected event. The neural network evaluates the accumulated radar data 160 when the event is detected, or after accumulating sufficient further radar data following the detected event. Optionally, the neural network can continue monitoring / evaluating radar data beyond this time period until a certain degree of certainty is reached, i.e., whether the event shown by the radar data is a gravity-related natural disaster or not.

[0045] If the neural network confirms the event detected by the "real-time" algorithm, the initiated action can remain unchanged. Otherwise, if the neural network rejects the event detected by the "real-time" algorithm, the initiated action can be revised. This may include canceling or reversing the initiated action, and / or initiating a reverse action that counteracts these actions. For example, if the initiated action includes an alarm, the revised action might include allocating the alarm signal; if the initiated action includes switching a traffic signal to "red," the revised action might include turning off the traffic signal (or switching it to "green," "flashing yellow," etc.).

[0046] Optionally, if the neural network confirms the detected event (170), it can initiate additional actions (190). Specifically, these further actions may include those that are less time-sensitive but require more effort. For example, if the initiated action includes switching traffic signals to "red," then the additional action could include closing barriers to physically close the road. Furthermore, further actions may be based on more internal information obtained by the neural network from evaluating the data (160). For example, depending on the size and direction of the avalanche, additional alarms could be triggered in some locations, while existing alarms could be deactivated in others.

[0047] Optionally, additional sensors can be integrated into the proposed solution. Specifically, the neural network can evaluate data from sources other than the radar data used by the "real-time" algorithm. These additional sensors can include seismic or infrasound sensors, cameras, microphones, and / or weather information to supplement the radar data. This provides a multimodal detection approach that improves system reliability and reduces false alarms through cross-validation events.

[0048] Although the invention has been described above with reference to some preferred embodiments, it should be understood that various modifications and combinations of features of the embodiments are possible. All such modifications fall within the scope of the appended claims.

Claims

1. A computer-implemented method (100) for detecting gravity-type natural disasters (1), the method comprising: - Real-time continuous reception of (110) radar data; - Using detection algorithm (5), continuously monitor (120) the radar data to detect (130) an indication of a gravity-type natural disaster event (1) in the radar data; and, - If an indication of a gravity-type natural disaster event (1) is detected (130), then action (140) is initiated in real time. Its features are, - Upon detecting (130) indication of a gravity-type natural disaster event (1), the trained neural network (6) evaluates (160) the radar data accumulated (150) for at least a limited time period; - The trained neural network (6) confirms (170) or refutes the gravity-type natural disaster event (1) based on the assessment (160); and - If the gravity-type natural disaster event (1) is refuted, the action that has been initiated is automatically revised (180).

2. The method (100) according to claim 1, wherein, The gravity-type natural disaster event (1) is one of avalanches, landslides, mudslides, debris flows, and rockfalls.

3. The method (100) according to claim 1 or 2, further comprising accumulating (150) the received radar data, particularly wherein, The received radar data is stored in a data memory and retrieved from the data memory for use in the evaluation (160).

4. The method (100) according to claim 3, wherein, The accumulation of radar data (150) begins when an indication of gravity-type natural disaster event (1) (130) is detected, and in particular, the defined time period begins from the time point at which the indication of gravity-type natural disaster event (1) is detected.

5. The method (100) according to claim 3, wherein, The accumulated (150) radar data includes the continuously accumulated (150) radar data, and in particular, the defined time period begins at a point in time before the detection (130) of the indication of the gravity-type natural disaster event (1) and ends at a point in time after the detection (130) of the indication of the gravity-type natural disaster event (1).

6. The method (100) according to any one of the preceding claims, wherein, The radar data is received (110) from a radar system (2) configured as a Doppler radar.

7. The method (100) according to any one of the preceding claims, wherein, The detection algorithm (5) is a rule-based algorithm.

8. The method (100) according to any one of the preceding claims, wherein, The action to initiate (140) includes: - Send commands to the alarm infrastructure (4), and / or - One or more components of a remote control alarm infrastructure (4), In particular, the alarm infrastructure (4) includes: - One or more traffic lights and / or barriers, wherein activating the action (140) includes controlling the traffic lights and / or barriers to close a road or railway track, - One or more alarms, wherein activating (140) the action includes controlling the alarm to trigger an audible emergency signal, and / or - One or more warning light signals, wherein activating (140) the action includes controlling the warning light to trigger a visual emergency signal.

9. The method (100) according to any one of the preceding claims, wherein, If the occurrence of the gravity-type natural disaster (1) has been confirmed (170), the method includes automatically initiating (190) further actions, particularly wherein initiating (190) the further actions includes: - Send commands to the alarm infrastructure (4), - One or more components of a remote control alarm infrastructure (4), and / or - Send one or more messages to personal devices located in areas potentially affected by the gravity-type natural disaster (1), particularly via one of GSM, Wi-Fi, fiber optic, radio, satellite communication and / or LoRa-WAN.

10. The method (100) according to any one of the preceding claims, wherein, The trained neural network (6) also evaluates (160) further data from one or more other sensors, particularly in which: - The other sensors include one or more of a seismic sensor, an infrasound sensor, a camera, an infrared sensor, and a microphone; - The further data was received upon detection of an indication of the gravity-type natural disaster event (1) as described in (130); - The one or more other sensors are activated upon detecting (130) an indication of the gravity-based natural disaster event (1); and / or - The trained neural network (6) also evaluates (160) weather data.

11. The method (100) according to any one of the preceding claims, wherein, The method is executed in a computing unit (3) having at least one first data connection for receiving (110) radar data from the radar system (2) in real time and continuously. In particular, the computing unit (3) includes at least one second data connection for sending commands to the alarm infrastructure (4) to initiate (140) the action and to revise (180) the initiated action.

12. A system for detecting gravity-type natural disasters (1), the system comprising a computing unit (3), the computing unit (3): - It has at least one first data connection for continuously receiving (110) real-time radar data from the radar system (2); and - A detection algorithm (5) is stored, which is configured to continuously monitor (120) the radar data to detect (130) an indication of a gravity-type natural disaster event (1) in the radar data; Its features are, The computing unit (3) stores a trained neural network (6) configured to, upon detecting (130) an indication of a gravity-type natural disaster event (1), assess (160) radar data accumulated (150) for at least a defined time period, and based on the assessment (160), confirm (170) or refute the gravity-type natural disaster event (1).

13. The system according to claim 12, wherein the system includes the radar system (2), and particularly wherein, The radar system (2) is configured as a Doppler radar.

14. The system of claim 12 or 13, wherein the system is configured to perform the method of any one of claims 1 to 11.

15. A computer program product comprising program code stored on a machine-readable medium or embodied by an electromagnetic wave containing program code segments, and having computer-executable instructions for performing the method (100) according to any one of claims 1 to 11 when executed, particularly on a computing unit (3) of a system according to any one of claims 12 to 14.

Images