Fire hydrant intelligent monitoring and protection device with AI image recognition

Abstract

The application discloses a fire hydrant intelligent monitoring and protection device with AI image recognition, which comprises an image acquisition module, an AI processing module, a state determination module, a communication control module and a power supply management module. The device collects the images of the fire hydrant and the surrounding area through wide-angle visible light and infrared thermal imaging, uses an embedded multi-class convolutional neural network model to identify the shielding, illegal opening and structural damage in real time, and generates a state code; in the abnormal state, an audible and light warning is triggered, information is reported through Bluetooth or NB-IoT, and long-term autonomous operation is realized in combination with solar energy and lithium batteries. The application can realize all-weather, intelligent and accurate monitoring and protection of the fire hydrant.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring of fire protection facilities and artificial intelligence image recognition technology, and in particular to an intelligent monitoring and protection device for fire hydrants with AI image recognition. Background Technology

[0002] Fire hydrants are critical facilities in urban fire protection systems, providing water for firefighting and playing a vital role in public safety and emergency rescue. Current fire hydrant monitoring methods largely rely on periodic manual inspections or simple mechanical sensors, which struggle to identify abnormal conditions such as obstruction, damage, or unauthorized opening in real time. Furthermore, they lack intelligent environmental sensing capabilities, resulting in delayed responses, low management efficiency, and failing to meet the real-time and precision requirements of smart fire protection. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent monitoring and protection device for fire hydrants with AI image recognition, which solves the problems mentioned in the background art.

[0004] This invention is implemented as follows: an intelligent monitoring and protection device for fire hydrants with AI image recognition, comprising: an image acquisition module, an AI processing module, a status determination module, a communication control module, and a power supply management module; the output end of the image acquisition module is connected to the input end of the AI ​​processing module; the image acquisition module is configured to continuously acquire image data of the fire hydrant body and its surrounding area; the output end of the AI ​​processing module is connected to the input end of the status determination module; the AI ​​processing module embeds a trained convolutional neural network model for feature extraction and classification of the image data output by the image acquisition module; the output end of the status determination module is connected to the input end of the communication control module; the status determination module generates a corresponding status code signal based on the recognition result output by the AI ​​processing module; the power supply management module is connected to the power supply ends of the image acquisition module, AI processing module, status determination module, and communication control module respectively, for providing a stable operating voltage and managing a low-power operation mode.

[0005] In an exemplary embodiment of the present invention, the image acquisition module includes: a wide-angle visible light camera and an infrared thermal imaging sensor; the wide-angle visible light camera has a field of view of not less than 120 degrees, is installed on the outer side of the top of the fire hydrant, and its lens faces the ground and the surrounding area; the infrared thermal imaging sensor is installed in parallel with the wide-angle visible light camera, and its detection surface is coaxially arranged with the visible light lens; the output terminals of both the wide-angle visible light camera and the infrared thermal imaging sensor are connected to the image input interface of the AI ​​processing module.

[0006] In an exemplary embodiment of the present invention, the AI ​​processing module includes: a dedicated AI inference chip, a storage unit, and an image preprocessing unit; the dedicated AI inference chip is connected to the image preprocessing unit via a PCIe interface, the image preprocessing unit receives raw image data from the image acquisition module, and performs denoising, white balance correction, and resolution normalization processing; the storage unit uses non-volatile flash memory and internally stores a pre-trained multi-class target detection model, which includes an occlusion recognition sub-model, an illegal opening behavior recognition sub-model, and a structural damage recognition sub-model; the dedicated AI inference chip calls the model parameters in the storage unit to perform forward inference operations on the preprocessed image data, and outputs a structured recognition result including target category, confidence level, and bounding box coordinates.

[0007] In an exemplary embodiment of the present invention, the state determination module includes: a logic judgment circuit and a state encoder; the input terminal of the logic judgment circuit receives the structured recognition result output by the AI ​​processing module, and is internally equipped with a threshold comparison unit. When the confidence level output by the occlusion recognition sub-model is greater than 0.85, or the illegal opening behavior recognition sub-model detects a hand operation sequence, or the structural damage recognition sub-model marks a crack area exceeding a preset pixel threshold, an abnormal state flag bit is triggered; the state encoder generates a four-bit binary state code based on the combination of abnormal state flag bits, corresponding to four state types: "normal", "occlusion", "illegal opening", and "structural damage"; the output terminal of the state encoder is connected to the data input terminal of the communication control module through a UART interface.

[0008] In an exemplary embodiment of the present invention, the communication control module includes: a short-range wireless communication unit and a long-range communication unit; the short-range wireless communication unit adopts the Bluetooth 5.2 protocol, and its antenna port is connected to a ceramic patch antenna disposed on the outside of the fire hydrant box, for sending status codes and location information to the handheld terminal of the inspection personnel; the long-range communication unit adopts an NB-IoT communication module, and its SIM card slot has a built-in operator IoT card, which uploads the status code, timestamp, and device ID to the urban fire protection IoT platform through the CoAP protocol; the enable terminals of both the short-range wireless communication unit and the long-range communication unit are connected to the control signal output terminal of the status determination module, and the communication function is activated only when the status code is not "normal".

[0009] In an exemplary embodiment of the present invention, the power supply management module includes: a solar charging unit, a rechargeable lithium battery pack, and a power switching circuit; the solar charging unit is composed of a monocrystalline silicon photovoltaic panel, installed on the top plane of the fire hydrant, and its positive and negative output terminals are connected to the input port of the power switching circuit; the rechargeable lithium battery pack is a 3.7V lithium polymer battery with a capacity of not less than 5000mAh, and its charging and discharging ports are connected to the energy storage interface of the power switching circuit; the power switching circuit uses a TPS63020 DC-DC conversion chip, and when the light intensity sensor detects that the ambient illuminance is lower than 50 lux, it automatically switches to lithium battery power supply and shuts down the power rails of non-essential modules.

[0010] In an exemplary embodiment of the present invention, the device further includes: an audible and visual prompting module; the audible and visual prompting module includes a high-brightness LED bead and a piezoelectric buzzer, and is installed in a conspicuous position on the front of the fire hydrant box; the anode of the high-brightness LED bead is connected to the light drive port of the status determination module through a current-limiting resistor, and the cathode is grounded; the signal terminal of the piezoelectric buzzer is connected to the audio drive port of the status determination module; when the status code is "illegal opening" or "structural damage", the status determination module outputs a PWM signal to drive the LED bead to flash at a frequency of 1Hz, and at the same time outputs a square wave signal to drive the piezoelectric buzzer to emit a warning sound of more than 85dB.

[0011] In an exemplary embodiment of the present invention, the device further includes: a geolocation module; the geolocation module employs a positioning chip supporting BeiDou / GPS dual-mode, and its antenna interface is connected to an active ceramic antenna integrated on the top of the fire hydrant; the serial output terminal of the geolocation module is connected to the auxiliary input terminal of the AI ​​processing module, which acquires the current latitude and longitude coordinates before each image acquisition is initiated, and embeds the coordinate information into the image metadata; the working power of the geolocation module is provided by the constant power rail of the power supply management module to ensure that the location information is continuously available.

[0012] The present invention provides an intelligent monitoring and protection device for fire hydrants with AI image recognition, offering the following advantages: The invention uses an image acquisition module to continuously monitor the fire hydrant and its surrounding environment; an AI processing module utilizes a dedicated inference chip to run a multi-task convolutional neural network model, enabling real-time identification of abnormal states such as obstruction, unauthorized opening, and structural damage; a state determination module generates precise state codes based on the identification results; a communication control module selects short-range or long-range communication methods to report abnormal information according to the state type; a power supply management module combines solar energy and lithium batteries to achieve long-term autonomous operation; an audible and visual alert module provides immediate warnings locally; and a geolocation module ensures the traceability of abnormal events. These modules work collaboratively through clear electrical connections and logical control sequences, solving the technical problem that traditional fire hydrant monitoring methods cannot identify various abnormal states in real-time, accurately, and autonomously, thus meeting the needs of intelligent fire protection systems for all-weather, intelligent monitoring of fire hydrant facilities. (See attached figures.) Figure 1 This is a schematic diagram of the overall block diagram of the intelligent monitoring and protection device for fire hydrants with AI image recognition according to the present invention; Figure 2 This is a control diagram of the intelligent monitoring and protection device for fire hydrants with AI image recognition according to the present invention.

[0013] Explanation of reference numerals in the attached diagram: 1. Fire hydrant body; 2. Wide-angle visible light camera; 3. Infrared thermal imaging sensor; 4. Solar charging unit; 5. Audio-visual prompt module; 6. Short-range wireless communication unit; 7. Long-range communication unit; 8. Geolocation module; 9. Power management module; 10. AI processing module. Detailed implementation method. This invention provides an intelligent monitoring and protection device for fire hydrants with AI image recognition, as detailed below. Figure 1 and attached Figure 2 Specific embodiments of the present invention will be described in detail below. For example... Figure 1 As shown, the device is integrated into the external structure of the fire hydrant body 1. The functional modules are installed and interconnected according to the spatial layout and electrical connection relationship to realize all-weather, intelligent monitoring and abnormal status response of the fire hydrant body 1 and its surrounding area. Figure 2 This is a top-view structural diagram of the device, further illustrating the spatial arrangement of the top components.

[0014] In this embodiment, the device includes an image acquisition module, an AI processing module 10, a status determination module, a communication control module, a power management module 9, an audio-visual prompting module 5, and a geolocation module 8. The image acquisition module consists of a wide-angle visible light camera 2 and an infrared thermal imaging sensor 3, which are mounted side-by-side on the top outer side of the fire hydrant body 1, with their lenses facing the ground and surrounding area. Their detection surfaces are coaxially aligned to ensure high consistency in spatial coordinates between the images acquired by both. The wide-angle visible light camera 2 has a field of view of no less than 120 degrees, used to cover the typical operating areas in front of and to the sides of the fire hydrant body 1; the infrared thermal imaging sensor 3 is used to acquire thermal distribution images in low-light or nighttime environments to assist in identifying personnel approaching or engaging in illegal operations. The outputs of both the wide-angle visible light camera 2 and the infrared thermal imaging sensor 3 are connected to the image input interface of the AI ​​processing module 10 via flexible cables to transmit raw image data.

[0015] The AI ​​processing module 10 is fixedly installed inside the waterproof cavity of the fire hydrant body 1. It includes a dedicated AI inference chip, a storage unit, and an image preprocessing unit. The image preprocessing unit is connected to the dedicated AI inference chip via a PCIe interface. It receives raw image data from the wide-angle visible light camera 2 and the infrared thermal imaging sensor 3, and performs noise reduction, white balance correction, and resolution normalization to unify images from different modalities into a standard format. The storage unit uses non-volatile flash memory and pre-programs a multi-category target detection model. This model includes three sub-models: an obstruction recognition sub-model, an illegal opening behavior recognition sub-model, and a structural damage recognition sub-model. After receiving the preprocessed image, the dedicated AI inference chip calls the corresponding model parameters in the storage unit, performs forward inference operations, and outputs a structured recognition result. This result includes the target category label, confidence score, and bounding box coordinates, and is transmitted to the status determination module via an internal bus.

[0016] The status determination module is also integrated into the waterproof cavity where the AI ​​processing module 10 is located. Its hardware consists of a logic judgment circuit and a status encoder. The input of the logic judgment circuit receives the structured recognition result output by the AI ​​processing module 10 and has an internal threshold comparison unit. When the confidence level output by the occlusion recognition sub-model is greater than 0.85, it is determined that there is a valid occlusion; when the illegal opening behavior recognition sub-model detects a continuous sequence of hand operation actions (such as wrench rotation, valve rotation, etc.), it is determined that there is an illegal opening behavior; when the structural damage recognition sub-model marks a crack area exceeding a preset pixel threshold (e.g., 200 pixels), it is determined that there is structural damage. When any of the above conditions are met, the logic judgment circuit triggers the corresponding abnormal status flag bit. The status encoder generates a four-bit binary status code based on the combination of abnormal status flag bits: "0000" indicates normal, "0001" indicates occlusion, "0010" indicates illegal opening, and "0011" indicates structural damage. The output of the status encoder is connected to the data input of the communication control module via a UART interface.

[0017] The communication control module includes a short-range wireless communication unit 6 and a long-range communication unit 7, both installed in recessed slots on the side wall of the fire hydrant body 1. The short-range wireless communication unit 6 uses a Bluetooth 5.2 protocol chip, and its antenna port is connected to a ceramic patch antenna located on the outside of the fire hydrant box. It automatically broadcasts the current status code and device location information when inspection personnel carrying handheld terminals enter within a 10-meter range. The long-range communication unit 7 uses an NB-IoT communication module, with a built-in carrier IoT card in its SIM card slot. It encapsulates the status code, timestamp, and device ID into data packets via the CoAP protocol and uploads them to the city's fire protection IoT platform via a cellular network. The enable terminals of both the short-range wireless communication unit 6 and the long-range communication unit 7 are connected to the control signal output terminal of the status determination module. The status determination module outputs a high-level signal to activate the communication function only when the status code is not "0000" (i.e., an abnormal state); otherwise, it remains in a low-power sleep state.

[0018] The power management module 9 is located in an independent power compartment at the bottom of the fire hydrant body 1, and consists of a solar charging unit 4, a rechargeable lithium battery pack, and a power switching circuit. The solar charging unit 4 is made of monocrystalline silicon photovoltaic panels and is installed on the top plane of the fire hydrant body 1, adjacent to the wide-angle visible light camera 2 and the infrared thermal imaging sensor 3. Its positive and negative output terminals are connected to the input port of the power switching circuit via waterproof connectors. The rechargeable lithium battery pack is a 3.7V lithium polymer battery with a capacity of 5000mAh, and its charging and discharging ports are connected to the energy storage interface of the power switching circuit. The power switching circuit uses a TPS63020 DC-DC converter chip. Its input terminal is connected to both the solar charging unit 4 and the rechargeable lithium battery pack, and its output terminal provides a stable operating voltage of 3.3V or 5V to the image acquisition module, AI processing module 10, status determination module, communication control module, audio-visual prompt module 5, and geolocation module 8, respectively. In addition, the power management module 9 also integrates a light intensity sensor, which detects the ambient illuminance in real time. When the illuminance is below 50 lux, the power switching circuit automatically cuts off the solar input path and switches to lithium battery power supply. It also notifies the AI ​​processing module 10 to shut down the power rails of non-essential modules and enter a low-power operation mode via GPIO signal.

[0019] The audible and visual warning module 5 is installed in a prominent position on the front of the fire hydrant body 1, and consists of a high-brightness LED bead and a piezoelectric buzzer. The anode of the high-brightness LED bead is connected to the light drive port of the status determination module through a 220Ω current-limiting resistor, and the cathode is grounded; the signal terminal of the piezoelectric buzzer is directly connected to the audio drive port of the status determination module. When the status code is "0010" (illegal opening) or "0011" (structural damage), the status determination module outputs a PWM signal with a frequency of 1Hz and a duty cycle of 50% to the light drive port, driving the LED bead to flash at a frequency of 1Hz; at the same time, it outputs a square wave signal with a frequency of 2.5kHz and an amplitude of 3.3V to the audio drive port, driving the piezoelectric buzzer to emit a warning sound with a sound pressure level of not less than 85dB, lasting for 30 seconds or until the status returns to normal.

[0020] The geolocation module 8 uses a positioning chip supporting both BeiDou and GPS dual-mode. Its package is fixed to the inner top of the fire hydrant body 1. The antenna interface is connected to the active ceramic antenna integrated on the top via a coaxial cable. This antenna, along with the wide-angle visible light camera 2, the infrared thermal imaging sensor 3, and the solar charging unit 4, is placed on the same mounting plane, but is physically isolated to avoid electromagnetic interference. The serial output terminal (UART_TX) of the geolocation module 8 is connected to the auxiliary input terminal (UART_RX) of the AI ​​processing module 10. Before each image acquisition is initiated, the AI ​​processing module 10 sends a positioning request command to the geolocation module 8. After responding, the geolocation module 8 returns the current latitude and longitude coordinates and positioning accuracy information. The AI ​​processing module 10 embeds this coordinate information into the EXIF ​​metadata of the acquired image and transmits it to the status determination module along with the structured recognition results. Finally, the location information is included when the communication control module uploads data, ensuring the geographical traceability of abnormal events.

[0021] During actual operation, after the device is powered on, the power management module 9 first detects the ambient light intensity. If the illuminance is higher than 50 lux, the solar charging unit 4 will provide power first, while simultaneously charging the rechargeable lithium battery pack; if it is lower than 50 lux, the power supply will switch to the lithium battery. After the AI ​​processing module 10 is initialized, the geolocation module 8 will be activated to obtain the current location coordinates, and the image acquisition module will be triggered to start working. The wide-angle visible light camera 2 will acquire visible light images at a frequency of one frame every 5 seconds, and the infrared thermal imaging sensor 3 will simultaneously acquire thermal imaging images. After the two images are processed by the image preprocessing unit, they will be input into the dedicated AI inference chip for multi-task inference. The inference result will be sent to the status determination module for logical judgment. If there is no abnormality, the status code will be "0000", and the communication control module will remain in sleep mode; if an abnormality is detected, the corresponding status code will be generated, and the short-range wireless communication unit 6 and the long-range communication unit 7 will be activated. At the same time, if the abnormality type is illegal opening or structural damage, the audio-visual prompt module 5 will immediately activate the local warning. Inspection personnel can receive Bluetooth broadcast information via handheld terminals to quickly locate abnormal fire hydrants; the city's fire protection IoT platform receives data uploaded by NB-IoT in real time to achieve remote monitoring and dispatch.

[0022] Electrical connections between modules are achieved via traces on the PCB board or shielded cables. All external interfaces use IP68-rated waterproof connectors to ensure long-term stable operation in humid, rainy outdoor environments. The AI ​​processing module 10 communicates with the status determination module via high-speed UART at a baud rate of 115200bps to ensure real-time transmission of status information. The communication control module and the status determination module are connected via both GPIO and UART, enabling both communication enable and status data transmission. The power management module 9 provides independent power rails for all modules and uses MOSFET switches for on-demand power supply, minimizing static power consumption. The entire device's mechanical structure uses an aluminum alloy shell with an anodized finish, providing corrosion and impact resistance, making it suitable for various deployment scenarios such as urban roads, industrial parks, and residential areas.

[0023] Through the physical layout, electrical connection, and collaborative working mechanism of the above modules, this invention achieves continuous visual perception, intelligent analysis, and multi-level response of the fire hydrant body 1 and its surrounding environment. This overcomes the technical shortcomings of traditional manual inspections, such as low efficiency, high false alarm rate of mechanical sensors, and inability to identify complex abnormal states, providing reliable technical support for intelligent fire protection systems. To better enable those skilled in the art to fully understand and implement this invention, the specific implementation principles are further supplemented below with a specific application scenario.

[0024] After the device is installed on the fire hydrant body 1 deployed next to the city road intersection, the ambient illuminance is first detected by the light intensity sensor in the power supply management module 9. When the ambient illuminance is higher than 50 lux during the day, the solar charging unit 4 converts light energy into electrical energy through the monocrystalline silicon photovoltaic panel and inputs it to the power switching circuit through the waterproof connector. This circuit uses a TPS63020 DC-DC converter chip to provide a stable 3.3V or 5V operating voltage for each functional module and to trickle charge the rechargeable lithium battery pack. When the illuminance is lower than 50 lux at night or in cloudy or rainy weather, the power switching circuit automatically disconnects the solar input path and switches to power supply by the 3.7V / 5000mAh lithium polymer battery. It also notifies the AI ​​processing module 10 through the GPIO signal to shut down the power rails of non-essential modules other than the infrared thermal imaging sensor 3, entering a low-power operation mode, thereby extending the battery life.

[0025] After the AI ​​processing module 10 completes initialization, it sends a positioning request command to the geolocation module 8 through the UART interface. The geolocation module 8 analyzes satellite signals based on the Beidou / GPS dual-mode positioning chip, obtains the current latitude and longitude coordinates and positioning accuracy information through the active ceramic antenna integrated on the top, and returns it to the UART_RX port of the AI ​​processing module 10 through the UART_TX port. The AI ​​processing module 10 writes the coordinate information into the EXIF ​​metadata field of the subsequently acquired images to ensure that each frame of the image carries accurate geographical location information.

[0026] Subsequently, the wide-angle visible light camera 2 starts image acquisition at a frequency of one frame every 5 seconds, with a field of view of no less than 120 degrees, covering an operating area of ​​approximately 3 meters × 3 meters in front of and to both sides of the fire hydrant body 1. At the same time, the infrared thermal imaging sensor 3 synchronously acquires thermal distribution images. The detection surfaces of the two sensors are set coaxially to ensure consistency of spatial coordinates. The raw image data is transmitted to the image preprocessing unit in the AI ​​processing module 10 via a flexible cable. This unit performs Gaussian noise reduction, white balance correction, and resolution normalization processing to unify the visible light and thermal imaging images into a standard format of 1080×1080 pixels, and then sends them to the dedicated AI inference chip through the PCIe interface.

[0027] The dedicated AI inference chip calls a pre-programmed multi-category target detection model from the storage unit. This model contains three independent sub-models: an occlusion recognition sub-model for identifying static obstacles such as vehicles, trash cans, and construction barriers; an illegal opening behavior recognition sub-model for identifying operation sequences such as wrench rotation and valve stem rotation by analyzing the relative motion trajectory of the hand and valve in consecutive frames; and a structural damage recognition sub-model for detecting whether there are cracks, corrosion, or missing parts on the surface of the fire hydrant body 1 based on edge enhancement and texture analysis algorithms. After inference, a structured result is output, including the target category label, confidence score (e.g., 0.92), and bounding box coordinates (e.g., x=150, y=200, w=80, h=60).

[0028] After receiving the structured results, the state determination module uses a threshold comparison unit within the logic judgment circuit to make a judgment: if the confidence level output by the occlusion recognition sub-model is greater than 0.85, the "occlusion" flag is set; if the illegal opening behavior recognition sub-model detects a sequence of actions matching the timing characteristics of hand operations for more than three consecutive frames, the "illegal opening" flag is set; if the crack area calculated by the structural damage recognition sub-model exceeds 200 pixels, the "structural damage" flag is set. The state encoder generates a four-bit binary state code based on the flag combination; for example, it outputs "0010" when only the "illegal opening" flag is set.

[0029] When the status code is not "0000", the status determination module outputs a high-level signal via GPIO, activating the short-range wireless communication unit 6 and the long-range communication unit 7. The short-range wireless communication unit 6 uses a Bluetooth 5.2 protocol chip and broadcasts a broadcast packet containing the status code, device ID, and geographic coordinates via an external ceramic patch antenna. The inspection personnel's handheld terminal can automatically receive and display the anomaly type and location within a 10-meter range. The long-range communication unit 7, through the NB-IoT module, uploads the encapsulated CoAP data packet (containing status code, timestamp, device ID, latitude and longitude) to the city's fire protection IoT platform via the cellular network to achieve remote alarm.

[0030] Meanwhile, if the status code is "0010" or "0011", the status determination module immediately outputs two drive signals: one is a PWM signal with a frequency of 1Hz and a duty cycle of 50%, which drives the high-brightness LED beads in the sound and light prompt module 5 to flash at a frequency of 1Hz through a 220Ω current-limiting resistor; the other is a square wave signal with a frequency of 2.5kHz and an amplitude of 3.3V, which directly drives the piezoelectric buzzer to emit a warning sound with a sound pressure level of not less than 85dB, lasting for 30 seconds or until the status returns to normal, thereby forming an immediate deterrent against illegal operation or serious damage on site.

[0031] The entire device's mechanical structure uses an aluminum alloy shell with an anodized surface. All external interfaces use IP68-rated waterproof connectors. The antennas of the wide-angle visible light camera 2, infrared thermal imaging sensor 3, solar charging unit 4, and geolocation module 8 are all located on the top plane of the fire hydrant body 1, but are electromagnetically isolated by a metal partition to prevent radio frequency interference from affecting positioning accuracy. The modules are connected via PCB traces or shielded cables. The AI ​​processing module 10 communicates with the status determination module via high-speed UART at a baud rate of 115200bps, ensuring that status information is transmitted within 50 milliseconds to meet real-time response requirements.

[0032] Through the coordinated execution of the above steps and principles, this invention can achieve all-weather, multi-modal, and high-precision abnormal state perception and hierarchical response of the fire hydrant body 1 and its surrounding environment under unattended conditions, effectively supporting the intelligent operation and maintenance needs of smart fire protection systems for infrastructure.

[0033] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A fire hydrant intelligent monitoring and protection device with AI image recognition, characterized in that, include: The system comprises an image acquisition module, an AI processing module (10), a status determination module, a communication control module, and a power management module (9). The output of the image acquisition module is connected to the input of the AI ​​processing module (10) for continuous image data acquisition of the fire hydrant body (1) and its surrounding area. The output of the AI ​​processing module (10) is connected to the input of the status determination module. The AI ​​processing module (10) has a trained convolutional neural network model embedded in it for feature extraction and classification of the image data output by the image acquisition module. The output of the status determination module is connected to the input of the communication control module for generating a corresponding status code signal based on the recognition result output by the AI ​​processing module (10). The power management module (9) is connected to the power supply terminals of the image acquisition module, the AI ​​processing module (10), the status determination module, and the communication control module, respectively, for providing a stable operating voltage and managing a low-power operation mode.

2. The intelligent monitoring and protection device for fire hydrants with AI image recognition as described in claim 1, characterized in that, The image acquisition module includes a wide-angle visible light camera (2) and an infrared thermal imaging sensor (3); the field of view of the wide-angle visible light camera (2) is not less than 120 degrees, and it is installed on the top outer side of the fire hydrant body (1), with the lens facing the ground and the surrounding area; the infrared thermal imaging sensor (3) is installed in parallel with the wide-angle visible light camera (2), and its detection surface is coaxial with the lens of the wide-angle visible light camera (2); the output ends of the wide-angle visible light camera (2) and the infrared thermal imaging sensor (3) are both connected to the image input interface of the AI ​​processing module (10).

3. The intelligent monitoring and protection device for fire hydrants with AI image recognition as described in claim 1, characterized in that, The AI ​​processing module (10) includes a dedicated AI inference chip, a storage unit, and an image preprocessing unit. The dedicated AI inference chip is connected to the image preprocessing unit via a PCIe interface. The image preprocessing unit receives raw image data from the image acquisition module and performs denoising, white balance correction, and resolution normalization. The storage unit uses non-volatile flash memory and stores a pre-trained multi-class target detection model, which includes an occlusion recognition sub-model, an illegal opening behavior recognition sub-model, and a structural damage recognition sub-model. The dedicated AI inference chip calls the model parameters in the storage unit to perform forward inference operations on the preprocessed image data and outputs a structured recognition result containing the target category, confidence level, and bounding box coordinates.

4. The intelligent monitoring and protection device for fire hydrants with AI image recognition as described in claim 1, characterized in that, The state determination module includes a logic judgment circuit and a state encoder; the input terminal of the logic judgment circuit receives the structured recognition result output by the AI ​​processing module (10), and is equipped with a threshold comparison unit. When the confidence level output by the occlusion recognition sub-model is greater than 0.85, or the illegal opening behavior recognition sub-model detects a hand operation sequence, or the structural damage recognition sub-model marks a crack area exceeding 200 pixels, the abnormal state flag is triggered. The status encoder generates a four-bit binary status code based on the combination of abnormal status flag bits, corresponding to four status types: "normal", "obstructed", "illegally opened" and "structural damage". The output of the status encoder is connected to the data input of the communication control module through a UART interface.

5. The intelligent monitoring and protection device for fire hydrants with AI image recognition as described in claim 1, characterized in that, The communication control module includes a short-range wireless communication unit (6) and a long-range communication unit (7). The short-range wireless communication unit (6) adopts the Bluetooth 5.2 protocol, and its antenna port is connected to a ceramic patch antenna set on the outside of the fire hydrant box, which is used to send status codes and location information to the handheld terminal of the inspection personnel. The long-range communication unit (7) adopts an NB-IoT communication module, and its SIM card slot has a built-in operator IoT card. It uploads the status code, timestamp and device ID to the urban fire protection IoT platform through the CoAP protocol. The enable terminals of the short-range wireless communication unit (6) and the long-range communication unit (7) are both connected to the control signal output terminal of the status determination module, and the communication function is activated only when the status code is not "normal".

6. The intelligent monitoring and protection device for fire hydrants with AI image recognition as described in claim 1, characterized in that, The power supply management module (9) includes a solar charging unit (4), a rechargeable lithium battery pack, and a power switching circuit. The solar charging unit (4) is composed of a monocrystalline silicon photovoltaic panel and is installed on the top plane of the fire hydrant body (1). Its positive and negative output terminals are connected to the input port of the power switching circuit. The rechargeable lithium battery pack is a 3.7V lithium polymer battery with a capacity of 5000mAh. Its charging and discharging ports are connected to the energy storage interface of the power switching circuit. The power switching circuit uses a TPS63020DC-DC converter chip. When the light intensity sensor detects that the ambient illuminance is lower than 50 lux, it automatically switches to lithium battery power supply and shuts down the power rails of non-essential modules.

7. The intelligent monitoring and protection device for fire hydrants with AI image recognition as described in claim 1, characterized in that, It also includes an audio-visual prompt module (5) and a geolocation module (8); the audio-visual prompt module (5) includes a high-brightness LED bead and a piezoelectric buzzer, which are installed in a prominent position on the front of the fire hydrant box; the anode of the high-brightness LED bead is connected to the light drive port of the status determination module through a 220Ω current-limiting resistor, and the cathode is grounded; the signal terminal of the piezoelectric buzzer is connected to the audio drive port of the status determination module. When the status code is "illegal opening" or "structural damage", the status determination module outputs a PWM signal to drive the LED beads to flash at a frequency of 1Hz, and at the same time outputs a square wave signal to drive the piezoelectric buzzer to emit a warning sound with a sound pressure level of not less than 85dB; the geolocation module (8) adopts a positioning chip that supports Beidou / GPS dual mode, and its antenna interface is connected to the active ceramic antenna integrated on the top of the fire hydrant body (1); the serial output terminal of the geolocation module (8) is connected to the auxiliary input terminal of the AI ​​processing module (10), which obtains the current latitude and longitude coordinates before each image acquisition is started, and embeds the coordinate information into the image metadata; the working power of the geolocation module (8) is provided by the constant power rail of the power supply management module (9).

Images