Device and method for detecting, classifying and scaring away a flying animal

Abstract

The present invention relates to a device and a method for detecting, classifying, and scaring away a flying animal, and in particular a bird species or a bat. The invention primarily relates to a device for detecting and scaring away a flying animal, comprising at least means for acquiring data in a surveillance area, means for analyzing the acquired data, means for generating at least one scaring signal, and optionally means for regulating the operation of a structure within an activity zone. The invention finds particular application in the fields of aeronautics, agriculture, energy production (such as wind or solar power), environmental protection, and any area of ​​human activity carried out at sensitive sites.A device and a method according to the invention are particularly useful for the detection and scaring away of birds when they enter a critical zone located around sites of human activity.

Description

Title of the invention: Device and method for detecting, classifying and scaring away a flying animal

[0001] The present invention relates to a device and a method for detecting, classifying, and scaring away a flying animal. A device and a method according to the invention are particularly useful for detecting and scaring away birds when they enter a critical zone located around sites of human activity.

[0002] The present invention therefore relates to the field of animal deterrent systems, and in particular to animal deterrent systems for birds. The invention finds particular application in the aeronautical, agricultural, and energy production sectors, such as wind and photovoltaic energy, as well as in environmental protection and any field of human activity carried out at sensitive installation sites.

[0003] Certain ecosystems of human activity, such as airports, offer birds a protected natural space and serve as a stopover for migratory species. However, collisions between birds and the structures of these activity sites generate significant bird mortality and cause material damage and / or a reduction in the site's functionality. Species posing a danger to these installations include birds of prey and species with gregarious behavior. Flying animals around sites of human activity can be injured or killed by these activities: avian species can be injured or killed in collisions with the blades of operating wind turbines, and bats can be injured by the strong pressure differences existing around operating wind turbines.

[0004] Bird risk prevention consists of reducing the presence of birds in the vicinity of sensitive activity sites. Bird scaring takes various forms such as driven hunts, regulated hunting, the use of audible or visual alarms, or the rearing of natural predators that act as deterrents near these areas.

[0005] EP 2 779 827 B1 describes a method and a device for scaring animals away from areas sensitive to human activity. The scaring method comprises the generation of at least one looming visual signal, this signal comprising at least one image of variable size over time.

[0006] WO 2014 / 085328 describes a system comprising a camera, detecting the presence and proximity of a bird in a defined area, and a means for emitting ultrasound or projecting substances intended to influence the flight path of a bird.

[0007] US 2013 / 0050400 describes a device for preventing bird collisions with wind turbines, in which at least one panoramic camera installed directly on a wind turbine generates images of its environment, and an evaluation system detects the presence of birds in that environment.

[0008] US 5,774,088 describes a device comprising the emission of microwaves intended to alert birds flying near areas of human activity. The emissions are detected by the birds' hearing and alert them to the presence of a protected area. The system can remain in standby mode until an antenna detects a bird and activates the microwave emission system.

[0009] In the presence of protected species, local legislation may require regulation of activity operations, and slowdowns and shutdowns result in a loss of productivity. It is therefore desirable to have means in place to protect birdlife while optimizing the operation and use of human activity sites located near bird habitats or migratory routes.

[0010] The present invention aims to overcome the aforementioned drawbacks. It relates to a device and a method for detecting and deterring flying animals, particularly avian species, near an activity site. Where applicable, the present invention ensures the regulation of the operation of structures at an activity site based on the risk of collision previously assessed by the device. BRIEF DESCRIPTION OF THE INVENTION

[0011] The inventors have developed a device and a method for the detection, classification and scaring away of a flying animal when it flies near a sensitive site of human activity, as well as the management of the operation of at least one structure of said site of activity.

[0012] A device and a method according to the invention have the advantage of providing an environmentally friendly and non-lethal solution that respects the environment and wildlife. The effectiveness of this device and method is immediate and lasting. Furthermore, they are easy to install and use.

[0013] Thanks to its analytical means, a device according to the invention allows, where appropriate, a precise identification of avian species and an analysis of the behavior of the avifauna, the data are analyzed in a flexible and global way, the efficiency of the systems can be identified, and corrections or adjustments can be made if necessary.

[0014] In the context of use in the vicinity of a wind farm, a device and a method according to the invention meet the requirements for measurement, monitoring, and analysis during the various phases of the wind farm project, from its design to its operation. They provide a reliable and scalable solution that meets increasing regulatory constraints and biodiversity protection requirements.

[0015] A device and a method according to the invention have demonstrated their high effectiveness in scaring away birds of prey flying near wind farms. By combining an effective detection and scaring system with optimized management of site operations, in which the operation of structures is modified only when necessary after scaring, a device and a method according to the invention protect birdlife while limiting disruptions to operations and losses resulting from these disruptions.

[0016] The invention has as its first object a device for detecting and scaring away a flying animal comprising at least: means for acquiring data in a surveillance area, means for analyzing the acquired data, means for generating at least one scaring signal and optionally means for regulating the operation of a structure in an activity area.

[0017] The invention has as its second object a method for detecting and scaring away a flying animal comprising at least the following steps: acquisition of data relating to any flying object in a surveillance area, analysis of the data acquired and, where appropriate, generation of at least one scaring signal and optionally the activation of the regulation of the operation of a structure in an activity area.

[0018] The invention also relates to a classification model, previously trained on a training dataset, for the detection and characterization of a flying animal and, where applicable, for generating means of scaring away a flying animal and optionally for regulating the operation of a structure in an activity zone. The invention relates to the use of a device, method, or classification model according to the invention for the detection and scaring away of a flying animal. Finally, the invention relates to a structure comprising a device according to the invention.

[0019] The objects and features of the invention will become clear upon reading the detailed description thereof, the examples and figures whose legend is described.

[0020] In the description of the present invention, the term "for" as in the expression "for detection" means "configured for detection".

[0021] When an interval or range of values ​​is indicated, the cited bounds are considered to be part of the interval or range of values. DETAILED DESCRIPTION OF THE INVENTION

[0022] According to a first object, the invention relates to a device for detecting and scaring away a flying animal, said device comprising at least: a) means for acquiring data relating to a surveillance zone located around an activity site, b) means for analyzing the data acquired during step a), said means being configured for at least one of the following actions: i) detect and identify flying objects present in the surveillance area, ii) classify the objects according to their nature and optionally according to their species, where appropriate classifying an object as a flying animal, iii) estimate the trajectory and speed of the objects, (iv) estimate the distance between the object and at least one structure at the activity site, (v) assess the risk of collision between the object and said structure and, if necessary, trigger an action, (c) means for generating at least one scaring signal, said at least one signal being managed by management means and emitted by transmitting means, and d) optionally, means for regulating the operation of at least one structure of the activity site, characterized in that the means for data acquisition, the means for data analysis, the means for generating at least one scaring signal, and optionally the means for regulating the operation of at least one structure are interconnected.

[0023] The term "flying animal" means any animal belonging to the class of vertebrates and capable of flight. This group includes, in particular, avian species, that is to say birds, belonging to the class "Aves", and chiropterans, or bats.

[0024] The invention relates particularly to a device for detecting and scaring away an animal classified among avian species.

[0025] The term "avian species" refers to a bird species as defined by the classification of birds by orders and families. Many avian species can be detected, identified, and deterred by a device according to the invention. Birds of prey, or raptors, are particularly likely to be detected by a device according to the invention. Among known raptors, one can notably mention: eagles, buzzards, harriers, falcons, sparrowhawks, kites, vultures, etc. Other birds can also be detected and deterred by the device according to the invention, such as corvids, seabirds such as gulls, terns, auks, skuas, and procellariiformes.

[0026] By "site of activity" is meant a sensitive site of human activity, in particular an airport facility, an onshore or offshore wind farm, a photovoltaic park, an agricultural area, an area of ​​installation of large electric pylons, and any area including buildings or structures that animals flying in the area could hit and / or damage during a collision.

[0027] The term "surveillance zone" refers to an airspace surrounding one or more activity structures that can be monitored by at least one data acquisition device. The dimensions of the surveillance zone depend on the location of the installation and on the technical capabilities and sizing of the data acquisition system and the bird-scaring signal used. For example, for a visual bird-scaring signal that emits at least one image, the size of the display panel is important: the larger the panels, the greater the detection range for birds. For example, with display panels measuring 4m x 4m, it is possible to cover an aircraft taxiway as well as the aircraft takeoff area.

[0028] The term "critical zone" means an airspace volume, included within the airspace volume of the surveillance zone, located around one or more structures of the activity site. The critical zone is the part of the surveillance zone in which the risk of a flying animal colliding with a structure is higher than anywhere else in the surveillance zone.

[0029] In the case of a wind farm, for example, the critical zone can be defined as a distance of 300 meters or less from at least one wind turbine. Depending on the performance of the data acquisition means of a device according to the invention, this distance can be increased. Generally, the critical zone includes an airspace extending between 300 and 1000 meters around an operational structure. Figure 1 represents a particular embodiment in which a 350-meter protective dome is located around a wind turbine.

[0030] By "regulation of the functioning of at least one structure of the site of human activity" is meant, for example: - for an airport: regulation of aircraft takeoff and landing, - for a wind farm: regulation of the operation of one or more wind turbines by decreasing, stopping, restarting and / or increasing activity.

[0031] The means for acquiring data relating to any flying object present in the surveillance area are configured to provide data concerning this area. These means are selected from: - Visual detection methods: such as a camera, which provides images and video data; this data comes from high-resolution cameras (including PTZ or thermal cameras), allowing continuous monitoring of the area, these methods enable the visual analysis of moving objects; - auditory recording systems, such as directional or omnidirectional microphones that capture sound data, this data is used to detect calls or other specific sounds emitted by birds, and - Radar-type detection methods measure precise parameters such as the position, speed and trajectory of detected objects; they are very effective for monitoring objects in low visibility conditions such as fog or night.

[0032] A device according to the invention comprises at least one, two, or three detection means, of a similar or different nature. For example, a device according to the invention comprises at least one camera, one microphone, and one radar. The combination of multiple detection means makes it possible to guarantee reliable multispectral data collection, covering all environmental conditions and ensuring redundancy between the sensors.

[0033] Said at least one detection means is installed so that data acquisition is optimal for monitoring the activity site. According to a particular embodiment, the acquisition means, for example microphones and thermal cameras for detecting bird and / or bat activity, are installed above the access door to a wind turbine for a ground-based study and in a nacelle for a study around the wind turbine rotor. Each of said detection means is connected to the means for analyzing the acquired data.

[0034] More specifically, a device according to the invention is configured for at least one of the following actions: - the detection and classification of animal species located in a surveillance zone, around a sensitive site of human activity, - the emission of at least one bird-scaring signal upon entry of the monitored bird into a critical area near said sensitive area of ​​human activity, and - the monitoring of the bird's trajectory during and after the emission of said bird-scaring signal and, optionally, the regulation of the function of at least one structure of said sensitive site of human activity.

[0035] Preferably, a device according to the invention is configured to perform at least the three aforementioned actions.

[0036] The means for analyzing data from a device according to the invention include, in particular, a system for recognizing images, sounds, and radar data. The means for data analysis are configured to process all image, sound, and / or radar data and perform the analysis of this data. In practice, the means for analyzing the data of a device according to the invention are configured to perform at least one of the following operations: - object detection, - object classification, - estimation of the trajectory and speed of objects, - estimation of the distance between an object and at least one structure of the activity site, - assessment of the risk of collision between the object and said structure and, if necessary, to trigger an action.

[0037] In practice, the means for analyzing the data of a device according to the invention are configured to perform one, two, three, four or all of the following operations: - object detection, - object classification, - estimation of the trajectory and speed of objects, - estimation of the distance between an object and at least one structure of the activity site, - evaluation of the risk of collision between the object and said structure and, where appropriate, to trigger an action.

[0038] “Object detection” means the processing of data collected by data acquisition means, using algorithms, to identify the object(s) present in the monitored area. The objective is to identify objects in real time while combining data from several sensors to maximize detection reliability.

[0039] By "classification of objects" is meant the categorization of objects according to their nature (for example: bird, drone, debris, or other) and, if possible, their species. For an object classified as an animal, the means for data analysis make it possible to identify the animal's species, in particular the avian species. This identification is based on at least one criterion such as: its size, its speed, its trajectory, and its behavior.

[0040] Estimating the trajectory and speed of the object aims to predict whether an object is approaching a structure dangerously and, if so, to estimate its time of arrival on said structure.

[0041] Estimating the distance between an object and at least one structure at the activity site involves, in particular, assessing whether the detected object is at a critical distance from the structure. Radar data combined with visual data allows for the calculation of the precise distance between the object and at least one structure. If the object is at a distance equal to or less than a threshold distance from a structure, the object has therefore crossed the threshold for entering the critical zone; the system considers this a risky situation and triggers an action. In other words, a flying object at a distance less than a threshold distance is considered to be in a critical zone around a structure or activity site. In the case of wind turbines, the threshold distance can be as low as 300 meters.

[0042] The means for data analysis are therefore configured to make a decision when a situation considered to be at risk occurs and to trigger an action intended to implement one or more appropriate measures to limit or avoid any collision.

[0043] The means for data analysis are further configured to send a signal to the means for generating at least one scaring signal when the animal crosses the entry threshold into a critical area of ​​the surveillance zone.

[0044] Optionally, the means for analyzing the data of a device according to the invention are configured to send a signal to the means for regulating the operation of the structure(s) at the activity site that the object is approaching. This signal is intended to reduce and / or stop the operation of the structure, urgently when necessary, if the result of the calculation indicates a risk of collision exceeding a predetermined value.

[0045] Optionally, depending on the nature of the structure or activity site, the action consists, in a second step, of sending a signal to the means of regulating the operation of the structure(s) of the activity site which the object is approaching.

[0046] According to a particular embodiment, the data analysis means of a device according to the invention are configured to calculate the probability of collision of the flying animal with at least one of the structures at the activity site. This calculation is based on the analysis of image, sound, and / or radar data, the analysis of its trajectory, and the estimation of its speed and the position of at least one structure at the activity site, with which the data analysis means have been previously programmed. The result of the collision probability calculation is then compared with a predetermined value programmed into the data analysis means. If the result of the calculation indicates a collision risk greater than a first determined value, a module regulating the operation of said structure is activated to reduce and / or stop the structure's operation.

[0047] According to a particular embodiment, the means for analyzing the data of a device according to the invention are configured to send an activation signal to a control module for the operation of said structure, to increase or restart its operation if the result of the calculation indicates a risk of collision lower than the determined value.

[0048] Finally, the means for analyzing the data of a device according to the invention are configured so that, when the flying object in question is located outside the critical zone, The system reverts to a standard monitoring phase, ready to detect and manage other potential threats. A device according to the invention thus makes it possible to ensure continuous monitoring of the activity site concerned.

[0049] According to a particular embodiment of a device according to the invention, the means for analyzing the data acquired by the detection means are based on a computer tool comprising a decision intelligence (DI, business intelligence or BI) module based on artificial intelligence (AI) technologies.

[0050] According to this particular embodiment, artificial intelligence performs complex analyses from the collected data and the business intelligence platform transforms the raw data into actionable information to assist in decision-making.

[0051] The data collected by the sensors are processed using advanced algorithms to identify objects present in the surveillance area. In specific embodiments: video data are analyzed by convolutional neural network (CNN) models such as YOLO (You Only Look Once) or SSD (Single Shot Detector); these algorithms quickly detect moving objects (birds, drones, or others); radar data are analyzed to identify objects based on their position, altitude, and movement; and a spectral analysis of the sound data is performed to recognize the sound signatures specific to bird species. Such a system makes it possible to identify objects in real time while combining data from several sensors to maximize detection reliability.

[0052] According to a particular embodiment, once objects have been detected, they are classified according to their nature and, if possible, their species. Data classification is performed using a convolutional neural network model: detected objects are categorized, for example, as a bird, drone, or debris. If specific models have been previously trained, the system can recognize precise species, particularly protected or endangered ones.

[0053] Where appropriate, sound signatures and radar-detected trajectories help to refine the classification.

[0054] This classification step makes it possible to prioritize objects requiring immediate intervention, in particular birds close to or belonging to sensitive species.

[0055] Estimating the trajectory and speed of detected objects is crucial for anticipating their behavior. Video and radar data are used to model the current trajectory and predict future movements using Kalman filters or other mathematical modeling methods. Radar sensors measure the speed objects are analyzed to assess the risks of impact or passage through critical areas. This step allows us to predict whether an object is approaching structures dangerously and to estimate its time of arrival.

[0056] Distance estimation and critical threshold verification are performed after estimating the object's trajectory and speed. Data analysis means assess whether the detected object is at a critical distance: Radar data combined with visual data allow the precise distance between the object and the structures to be calculated. If the object is at a distance less than a predetermined threshold, the system considers this a risky situation and triggers an action. More specifically, in a device according to the invention, said threshold can be set at 300 meters from the structures. This step aims to activate, if necessary, an appropriate signal to limit the risk of collision.

[0057] More particularly, in a device according to the invention, the means for data analysis have been previously trained with a relevant dataset.

[0058] Even more specifically, the means for the data analysis have been previously trained with a dataset relating to at least one avian species, in order to determine precisely whether the animal belongs to said avian species.

[0059] By "determination of the entry threshold into the critical zone" we mean the prior programming of the means of data analysis with the localization of the entry threshold into the critical zone, and the analysis of image, sound and / or radar data, according to the results of the data analysis, determination of the crossing by the animal of the entry threshold into the critical zone is carried out.

[0060] The means for data analysis are configured to send an activation signal to the means for generating at least one scare signal when the animal crosses the entry threshold into a critical area of ​​the surveillance zone.

[0061] According to a particular aspect, in a device according to the invention, the means for generating at least one scaring signal are managed by at least one management means and emitted by at least one emitting means.

[0062] The use of light stimuli such as flashing lights and laser beams is widely described. Sonic deterrents exploit the auditory sensitivity of birds by emitting different types of sounds: ultrasound, predator calls, distress calls, or disruptive artificial sounds.

[0063] In a device according to the invention, a scaring signal can be chosen from: a visual signal, in particular a light signal or a signal comprising at least the diffusion of an image, an auditory signal, in particular an ultrasound signal, or a combination of signals.

[0064] The means for generating said scaring signal are well known to a person skilled in the art.

[0065] Said transmission means are placed in locations selected for their visibility or accessibility to a flying animal within the surveillance area. These transmission means are placed in the critical zone. A person skilled in the art can determine the optimal installation locations based on the nature of the activity site. According to one particular embodiment, round LED screen-type light panels with a diameter of at least 1 meter are fixed to the mast of a wind turbine at a height of between 6 and 12 meters above the ground. According to another particular embodiment, light panels are installed at various strategic points within the activity area, depending on the layout of the activity site.

[0066] More particularly, the invention relates to a device for the visual detection and scaring away of a flying animal, according to which the scaring signal is a looming signal in which the size of at least one image varies over time, an increase in the size of the image simulating an approach by increasing the size of the image.

[0067] The looming effect is an optical illusion consisting of emitting the image of an object of varying size progressively. As the image size increases, the viewer has the illusion of an object that is rapidly growing, as if it were approaching the viewer at high speed. A signal with the looming effect simulates an imminent collision or attack for a bird and triggers its flight. The term "looming" means "imminent" or "appearing" in English (Hausberger et al. "Wide-eyed glare scares raptors: from laboratory evidence to applied management", PLOS One, October 11, 2018, https: / / doi.org / 10.1371 / journal.pone.0204802).

[0068] Observations made during the scaring of raptors show that the use of such a signal leads to a rapid change in the behavior of raptors and that this effect is lasting, even when the bird population is large locally (Hausberger et al, 2018).

[0069] In a device according to the invention, a looming effect can be achieved using an image representing black concentric circles on a white background, or an image symbolizing one or two eyes, by enlarging the image size. The animation of the concentric circles is designed to maximize the effect of rapid looming. Preferably, in a device according to the invention, the image used for the visual signal consists of a pair of black discs.

[0070] Birds of prey likely to be scared away by a looming signal include: kites, falcons, eagles, buzzards, harriers, sparrowhawks, vultures, etc.

[0071] According to a particular aspect, in a device according to the invention, the means for generating at least one visual scaring signal, in which said visual signal comprises at least one image, are managed by at least one management means and emitted by at least one emitting means.

[0072] The visual scaring signal includes at least the emission of an image, figurative or abstract, for a variable duration depending on the programming of the device according to the invention, according to a form and a sequence chosen in order to cause the scaring of at least one flying animal, depending on the species considered.

[0073] The means for emitting the visual scaring signal are chosen from: screens, such as LED screens.

[0074] Preferably, when said visual signal is generated, it is visible for a duration of between 5 and 10 minutes, depending on the presence detected in the critical zone. The increase in image size can be gradual, and the rate of increase can be regular or variable. The rate of increase is configured according to a fixed parameter or according to the estimate of the bird's approach speed or its distance from the structure of the activity site. The frequency of signal size increase is also configured according to a fixed parameter or according to the estimate of the bird's approach speed or its distance from the structure of the activity site.

[0075] A device according to the present invention comprises at least one means for generating a visual signal, this means being in particular chosen from: an electronic screen (or panel) which directly displays the visual signal and a projection system which projects the signal onto a chosen surface.

[0076] In a device according to the invention, the means for data acquisition, the means for data analysis, the means for generating at least one scaring signal, and optionally the means for regulating the operation of at least one structure, are interconnected.

[0077] The detection means and the analysis means are computer-based and are interconnected using RJ45 and / or HDMI cables in the case of LED screens.

[0078] According to a particular embodiment, the invention relates to a device for the visual detection and deterrence of a flying animal, said device comprising at least: a) means for acquiring data relating to any flying object in the surveillance zone located around an activity site, to detect the possible presence of at least one flying object, and to track the trajectory of any flying object in said surveillance zone, b) means for analyzing the data acquired during step a), said means being configured to: i) detect and identify objects present in the surveillance area, ii) classify the objects according to their nature and optionally according to their species, iii) estimate the trajectory and speed of the objects, (iv) estimate the distance between the object and at least one structure at the activity site, (v) assess the risk of collision between the object and said structure and, if necessary, trigger an action, (c) means for generating at least one visual scaring signal, comprising at least one image, said at least one visual signal being managed by management means and emitted by transmission means, a looming signal in which the size of at least one image varies over time, simulating an approach by increasing the size of the image, said at least one signal being managed by management means and emitted by transmission means, d) means for regulating the operation of at least one structure at the activity site, characterized in that the means for data acquisition, the means For data analysis, the means for generating at least one scaring signal, and optionally the means for regulating the operation of at least one structure, are interconnected.

[0079] According to this particular embodiment, a device according to the invention comprises means for regulating the activity of a mobile structure: slowing down, stopping, emergency stopping, resuming activity, and regulating speed after the animal has passed. This device is particularly suitable when said structure is a wind turbine.

[0080] In the latter case, depending on the result of the data evaluation, the means for data analysis are capable of sending a signal: i. to means of generating a visual deterrent signal and / or ii. to the means of regulating the operation of at least one structure of the activity site.

[0081] According to a second object, the invention relates to a method for detecting and scaring away a flying animal, this method comprising at least the following steps: a) the acquisition of data relating to any flying object in a surveillance zone located around an activity site, to detect the presence of at least one flying object, b) the analysis of the data acquired during step a), including at least one of the following steps: i) the detection and identification of objects present in the surveillance area, ii) the classification of objects according to their nature and optionally according to their species, iii) the estimation of the trajectory and speed of objects, (iv) estimating the distance between the object and at least one structure at the activity site, (v) the assessment of the risk of collision between the object and said structure and, where appropriate, the triggering of an action, c) where appropriate the generation of at least one scaring signal, said at least one signal being managed by management means and emitted by transmission means, d) optionally, the regulation of the operation of at least one structure of the activity site.

[0082] Figure 2 schematically illustrates a particular embodiment of a method according to the invention for monitoring a wind turbine, with the various steps of the method. These steps include, in particular: data acquisition, data analysis, activation of the deterrent system, activation of the wind turbine's operating control module, and continued monitoring.

[0083] A method for detecting and scaring away a flying animal according to the invention is characterized in that the acquisition of data relating to any flying object in the surveillance zone located around an activity site is carried out for the entire duration of the presence of the animal in the surveillance zone, and more particularly for the duration of the presence of the animal in the critical zone and at least until its exit from the critical zone.

[0084] More particularly, the invention relates to a method for detecting and scaring away an avian species.

[0085] More particularly, the invention relates to a method for visually detecting and scaring away an avian species, according to which the scaring signal is a looming signal in which the size of at least one image varies over time, an increase in the size of the image simulating an approach by increasing the size of the image.

[0086] According to one particular aspect, a method for detecting and scaring away a flying animal according to the invention further comprises: i) the calculation of the risk of collision of the flying animal detected during step b) with a structure at the activity site, ii) if the result of the calculation indicates a risk of collision greater than a first determined value, the activation of a module to regulate the operation of said structure, to reduce and / or stop the operation of the structure.

[0087] According to this particular embodiment, a method according to the invention is implemented during the monitoring of an activity site comprising wind turbines, the activity of which is regulated according to the probability of collision of a flying animal with at least one wind turbine.

[0088] According to one particular aspect, a method for detecting and scaring away a flying animal according to the invention further comprises, after the activation of a module regulating the operation of said structure, when the probability of collision of the flying animal with the structure is less than a second determined value, the regulation module of the operation of said structure is activated to increase and / or restart the operation of the structure.

[0089] More particularly, the invention relates to a method for visually detecting and scaring away an avian species, this method comprising at least the following steps: a) the acquisition of data relating to any flying object in a surveillance zone located around an activity site, to detect the presence of at least one flying object, b) the analysis of the data acquired during step a), to identify among said flying object the presence of at least one animal, and where appropriate assess its trajectory and speed in the surveillance zone, the calculation of a collision risk of the flying animal detected during step b) with a structure of the activity site, and, if the result of the calculation indicates a collision risk greater than a first determined value, the activation of a module regulating the operation of said structure, to reduce and / or stop the operation of the structure, c) the detection of the animal crossing the entry threshold into a critical zone of the surveillance area and the emission of at least one visual deterrent signal as soon as the entry threshold into the critical zone is crossed, in order to move the animal away from at least one structure of the activity site, d) monitoring the animal's trajectory within the critical zone, and e) After the activation of a control module for the operation of said structure, when the probability of collision of the flying animal with the structure is less than a second determined value, the control module for the operation of said structure is activated to increase and / or restart the operation of the structure.

[0090] According to a third object, the invention relates to a classification model, previously trained on a training dataset, configured to analyze input data provided by the detection means of a device according to the invention.

[0091] The term "classification model" refers to a pre-trained machine learning algorithm, particularly one trained through supervised learning, as well as a training dataset for training said algorithm and an evaluation dataset. A classification model may consist of a computer program, said computer program being able to be written in any suitable computer language known to a person skilled in the art. Said computer program is capable of being implemented on a computer to generate a technical result. Examples of these technical results are provided below.

[0092] The training dataset may include a training set and a test set of the model. The model can thus be tested against the test dataset, and the test set can be used to determine whether the model training is satisfactory. The training set and the test set may be different. Alternatively, the test set may correspond to a portion of the training set. The classification model according to the invention detects and identifies risks by implementing fine discrimination; the different types of observation (visual, auditory, and radar) are correlated, and the generated data sets are rapidly analyzed and utilized. Furthermore, the site configuration can be used to amplify the analytical power and improve risk qualification.

[0093] A classification model according to the invention is adapted for classification (among avifauna and within avian species...) based on at least one characteristic of said flying object, to analyze the input data provided by the detection means and classify a device according to the invention as either "aircraft" or "bird," as well as "which avian species," "which trajectory," and "which speed." Such a classification model offers savings in time, cost, and performance compared to existing tools on the market.

[0094] In the case of the classification model, the input data are chosen from: - images with objects, - sound recordings, - data from radar.

[0095] The output data are objects with a label: a percentage of belonging to a specific species.

[0096] The training dataset may include a multitude of pairs of data, each pair of data comprising a first data representing at least one characteristic of said flying object (bird) and a second data representing membership in a given avian species.

[0097] Said classification model can be implemented on a computer to generate a technical result consisting of the classification of an object, based on its characteristics, into a particular avian species. The classification model is considered to have achieved a satisfactory level of learning across all profiles in the test set if the classification achieves, in particular, a minimum Fl score of 70%, 75%, 80%, 85%, 90%, 91%, or 92%.

[0098] According to a particular embodiment of a device according to the invention, the data analysis means implement a Kalman algorithm. This algorithm makes it possible to process key functionalities such as: - Real-time tracking: estimation of object positions and speeds from noisy data and continuity of tracking even in the event of temporary data interruptions, - Filtering of noisy data: smoothing of data from sensors (cameras, radars, lidars) to generate coherent trajectories, - Trajectory prediction: anticipating movements to predict future positions and application in bird-scaring and PTZ camera adjustment systems, - Multi-sensor data fusion: combining data from multiple sensors to improve accuracy, - Anomaly detection: identification of unusual behaviors, such as a sudden change of direction, - Reduction of false positives: reduction of extraneous objects by modeling the typical movements of birds, - Automatic Controls: triggering specific actions, such as scaring away or stopping wind turbines.

[0099] In one embodiment of a device according to the invention, a classification model is combined with a Kalman algorithm to improve real-time detection and tracking. This coupling makes it possible to process noisy data and anticipate the movements of detected objects. In an example of an AI and Kalman algorithm implementation of a device according to the invention, bird detection and tracking include: - Detection: a neural network detects birds in images / videos; the detected coordinates (x, y) are used as inputs for the Kalman algorithm. - Prediction and Update: inputs (Positions (x, y, z) and velocities with uncertainties), outputs (Corrected trajectories and future predictions), steps: prediction phase: (modeling of the movement to estimate the new positions) and update phase (adjustment of the predictions with the new measurements).

[0100] The invention also relates to a classification model, previously trained on a training dataset, to detect and classify, in a process according to the invention, at least one bird species or a bat.

[0101] According to a particular aspect, a classification model according to the invention meets at least one, at least two, or all of the following characteristics: - Model architecture based on a YOLO (You Only Look Once) type neural network - The model is supervised and requires at least 1500 images and annotations per category. - the main layers are convolutional (Conv2D) for feature extraction, - A "Batch Normalization" step leads to stabilizing the learning, - A "Dropout" step reduces overfitting, - a "Pooling Layers" step reduces dimensionality.

[0102] According to a particular aspect, a training algorithm for a classification model according to the invention meets at least one, at least two, or all of the following characteristics: - Optimizer: Adam (improved gradient descent), - Loss function: cross entropy, - Number of epochs: at least 30, adjusted according to observed performance. Overfitting should be avoided. - Performance and Validation Criteria: A minimum Fl score of 70% is required to consider the model as performing, preferably a classification model according to the invention achieves a minimum Fl score of 75%, 80%, 85%, 90%, 91% or 92%.

[0103] The predictions consist of probabilities of the detected objects belonging to specific classes.

[0104] According to a fourth object, the invention relates to the use of a device, method, or model according to the invention for detecting and scaring away a flying animal, in particular an animal of an avian species, in the vicinity of at least one site of human activity. A device, method, and model according to the invention are usable during the design, construction, and / or operation of an activity site.

[0105] More particularly, the invention relates to the use of a device, method, or model according to the invention for the detection and deterrence of avian species, particularly birds of prey, in the vicinity of a wind farm. Even more particularly, the invention relates to the use of a device, method, or model according to the invention in which the deterrence signal is a looming signal.

[0106] Furthermore, the invention relates to the use of a device, a method or a model according to the invention for evaluating the effectiveness of a scaring signal.

[0107] According to a fifth object, the invention relates to a fixed structure, in particular a wind turbine, characterized in that it comprises, or is connected to, a device for detecting and scaring away a flying animal, in particular an avian species and in particular a visual scaring device with a looming effect, according to the invention.

[0108] A wind turbine comprises a mast equipped with a nacelle fitted with an electric generator, said electric generator is driven in rotation by a rotor fitted with blades subjected to the action of the wind, the generator produces electricity when a shaft is driven in rotation.

[0109] The present invention will be better understood by reading the following examples, which are given to illustrate it and not to limit its scope.

[0110] Fig. 1 represents a critical zone represented by a protective dome with a radius of 350 meters around a wind turbine.

[0111] Figure [Fig.2] schematically illustrates a particular embodiment of a process according to the invention, with different successive stages of the process. EXAMPLES

[0112] Example 1: Device for detecting bird species near a wind farm

[0113] The inventors have developed and perfected an artificial intelligence-based bird detection and protection solution.

[0114] A detection and deterrent device according to the invention is configured for the detection of multiple objects and comprises the following detection means: visual detection (fixed camera or Pan Tilt Z (PTZ)), sound detection, and radar. The combination of different detection sources, for example, sound and visual, allows for the correlation of observations.

[0115] The surveillance area is defined by the installation of the bird monitoring and detection system on the mast of a wind turbine at a height between 6 and 15 m. The acquisition elements (microphones / camera and thermal cameras) are fixed to the structure of the wind turbine using neodymium magnets.

[0116] The images, sound data and radar data from the detectors are processed by the analysis means of the device, to which they are connected.

[0117] The analysis means include a technical decision intelligence module that uses all the data collected by the various sensors to analyze the movement of flying objects in the area monitored by the detectors. This technical module consists of artificial intelligence (AI) and a Business Intelligence (BI) platform.

[0118] The technical module includes, in particular: - a convolutional neural network (CNN) for multi-object detection configured for automatic visual presence detection in the surveillance area, - a convolutional neural classification network, trained for the detection of at least one avian species, configured for the assisted identification of at least one avian species chosen from: red kite, black kite, black stork, vultures, buzzards, seabirds, etc.....any species of flying animal is potentially identifiable according to the prior training of the classification model.

[0119] The infrastructure of a device according to the invention comprises a public cloud-based deployment, an on-premises server available with a Linux x86_x64 host and real-time multi-camera scenarios, and the possibility of deployment on sites with low or no connectivity.

[0120] The system components are computer-type and are interconnected using RJ45 and / or HDMI cables for the LED screens. All system components are powered by the wind turbine's electrical current. In the event of a lack of connectivity, remote monitoring systems provide continuous updates on any potential malfunctions. If necessary, contacting the site operator allows for the triggering of an emergency on-site intervention.

[0121] The tasks performed by the analytical means are: a) the detection of multiple objects, by implementing at least one of the means below, and preferably by combining the means below: - SSD / Yolo type convolutional neural network (CNN) for processing a streaming video source, - High-frequency image analysis, - Convolutional neural network (CNN) for localization based on radar data (Radar Detect), b) object classification using a convolutional neural network (CNN), c) trajectory detection and analysis using classical mathematical analysis, d) decision-making by means of a decision-making module established on the basis of rules set by the user.

[0122] The detection performance of a device according to the invention makes it possible to obtain: - the precise detection and counting of birds from ground cameras (fixed or Pan Tilt Z motorized tilting camera with zoom), - accurate bird classification, even in challenging environments, - accurate detection of small and large birds; the reference detection ranges for an AXIS Q1656-PLE camera at a resolution of 1920x1080 pixels are, for example, approximately 170 m for a wingspan of 0.68 m and approximately 400 m for a wingspan of 1.60 m; birds are detected down to 5x5 pixels.

[0123] A detection device according to the invention therefore allows for the identification, monitoring and analysis of the behavior of different bird species in real time and on a permanent basis.

[0124] The device makes it possible to identify, track, and analyze in real time the behavior of birds in environments such as wind farms. This device includes a decision-making intelligence module that combines two main technologies: - Artificial Intelligence (AI): This technology performs complex analyses based on the collected data, - Business Intelligence (BI): This platform transforms raw data into actionable information to support decision-making. This module uses data captured by various devices such as cameras, radar, and microphones to analyze bird movement and behavior.

[0125] Statistical analysis of observations of birds of prey and corvids shows that they are deterred by the visual stimulus without habituation. The deterrent effect is demonstrated by the decrease in the number of birds within the stimulus's visibility zone, despite an increase in the total number of birds throughout the site during the observation period.

[0126] Example 2: Device and method for detection and deterrence according to the invention in the vicinity of a wind farm

[0127] According to a particular embodiment of the invention, cameras, recorders and radar are installed near a wind farm on the mast of each wind turbine or on a delivery point present in the area.

[0128] These detection means are connected to analysis means as described in Example 1. The technical module includes, in particular, a convolutional neural network for classification, trained to detect several bird species, including the red kite. The critical zone is defined, in accordance with applicable regulations, at a distance greater than or equal to 300 meters from any wind turbine. The technical module is configured to detect the crossing of the entry threshold into the critical zone by at least one bird.

[0129] The infrastructure of the device further includes at least means for emitting a visual warning signal as soon as the bird crosses the entry threshold into the critical zone, the warning consisting of the emission of looming effect signals, and means for regulating the operation of the monitored structure.

[0130] The bird's trajectory and speed within the monitoring area are assessed. If necessary, a visual signal is emitted as soon as the bird enters the critical area. The looming effect is triggered. The bird's trajectory is constantly tracked as long as it remains within the critical zone.

[0131] If the bird has deviated from its trajectory after the emission of the visual scaring signal, its flight is followed until it leaves the critical area.

[0132] If the bird has not deviated sufficiently from its trajectory, the wind turbine's operating control system is activated: depending on the situation, the turbine's operation is slowed down or, more rarely, stopped. The bird's flight path is tracked until it leaves the critical zone.

[0133] In the event that the trajectory of the bird makes a collision with the blades of the wind turbine probable as soon as it enters the critical zone, the wind turbine's operating regulation device is activated in order to cause an emergency stop.

[0134] Statistical analysis of observations of birds of prey and corvids shows that they are deterred by the visual stimulus, without habituation. The deterrent effect is demonstrated by the decrease in the number of birds within the area of ​​visibility of the stimulus, despite an increase in the total number of birds throughout the site during the observation period.

Claims

Demands

1. A device for detecting and deterring a flying animal, said device comprising at least: a) means for acquiring data relating to a surveillance area located around an activity site, b) means for analyzing the data acquired during step a), said means being configured to perform at least one of the following actions: i) detecting and identifying flying objects present in the surveillance area, ii) classifying objects according to their nature and optionally according to their species, where appropriate classifying an object as a flying animal, iii) estimating the trajectory and speed of objects, iv) estimating the distance between the object and at least one structure of the activity site, v) assessing the risk of collision between the object and said structure and, where appropriate, triggering an action, c) means for generating at least one deterrence signal,said at least one signal being managed by management means and emitted by transmission means, and (d) optionally, means for regulating the operation of at least one structure of the activity site, characterized in that the means for data acquisition, the means for data analysis, the means for generating at least one scaring signal, and optionally the means for regulating the operation of at least one structure, are interconnected.

2. Device according to the preceding claim, characterized in that said flying animal is classified among avian species.

3. Device according to any one of claims 1 or 2, characterized in that said scaring signal is a looming visual signal in which the size of at least one image varies over time, simulating an approach by increasing the size of the image.

4. A method for the visual detection and deterrence of a flying animal, comprising at least the following steps: a) the acquisition of data relating to any flying object in a monitored area located around an activity site, in order to detect at least one flying object, b) the analysis of the data acquired during step a), comprising at least one of the steps for: i) the detection and identification of objects present in the monitored area, ii) the classification of objects according to their nature and optionally according to their species, iii) the estimation of the trajectory and speed of the objects, iv) the estimation of the distance between the object and at least one structure of the activity site, v) the assessment of the risk of collision between the object and said structure and, where appropriate, the triggering of an action, c) the generation of at least one deterrent signal, said at least one signal being managed by management means and emitted by transmission means, d) optionally, the regulation of the operation of at least one structure of the activity site.

5. Method according to claim 4, characterized in that said scaring signal is a visual signal with a looming effect, in which the size of at least one image varies over time, simulating an approach by increasing the size of the image.

6. A method according to any one of claims 4 or 5, further comprising: i) calculating the risk of collision of the flying animal detected during step b) with a structure of the activity site, ii) if the result of the calculation indicates a risk of collision greater than a first determined value, activating a module for regulating the operation of said structure, to reduce and / or stop the operation of the structure.

7. A method according to any one of claims 4 to 6, characterized in that it further comprises, after the activation of a control module for the operation of said structure, when the probability of collision of the flying animal with the structure is less than a second determined value, the control module for the operation of said structure is activated to increase and / or restart the operation of the structure.

8. Classification model for implementing a device according to any one of claims 1 to 3 or a method according to any one of claims 4 to 7.

9. Use of a device according to any one of claims 1 to 3, of a method according to any one of claims 4 to 7, or of a model according to claim 8 for the detection and scaring away of a flying animal.

10. Fixed structure, in particular wind turbine, characterized in that it is connected to a device for detecting and scaring away a flying animal according to any one of claims 1 to 3.

Images