Shared bicycle riding overload intelligent identification method, system, equipment and medium

Abstract

The invention discloses a shared bicycle riding overload intelligent identification method, system and device and a medium, and relates to the technical field of overload detection, and the method comprises the steps: collecting the temperature and pressure of a shared bicycle cushion; obtaining an original pressure diagram and an original temperature diagram; constructing input tensors of three channels through the original pressure diagram and the original temperature diagram; inputting the plurality of input tensors into a pre-trained convolutional neural network model for pattern recognition, and obtaining a classification result that the input image belongs to different riding scenes; according to the classification result, the shared bicycle is controlled to execute corresponding operation, and when the classification result is an overload scene, an alarm is triggered or running of the bicycle is limited. All complex intelligent analysis and decisions are completed at the cloud, and finally an overload judgment result is fed back to a vehicle to execute corresponding operation, so that an accurate, reliable and intelligent overload treatment effect is achieved.

Description

A method, system, device and medium for intelligent identification of overloaded shared bicycle riders. Technical Field

[0001] This invention relates to the field of overload detection technology, and in particular to a method, system, device and medium for intelligent identification of overloaded shared bicycles. Background Technology

[0002] In recent years, traffic congestion and environmental pollution during peak hours have become increasingly prominent problems in major cities. Shared electric bikes, as a new and environmentally friendly mode of transportation, are characterized by their low price and convenience, and can appropriately alleviate traffic congestion, thus gaining increasing popularity.

[0003] However, some riders disregard the dangers and carry passengers, posing a threat to traffic safety. To reduce the overloading of shared electric bikes, traffic regulatory departments in various regions have repeatedly issued safety warnings, emphasizing the need to strengthen the management of such illegal activities. However, due to the difficulty of supervision and the weak traffic safety awareness of some citizens, this phenomenon persists despite repeated crackdowns.

[0004] Taking the shared electric bicycle market in Nanchang as an example, the existing overload detection technology is still in its early stages. Its technical form and actual application effect have obvious defects and cannot effectively deal with the widespread overload problem.

[0005] Current technological approach: Transformer analysis based on "gravity sensing". Currently, technological attempts by major operators like Meituan primarily rely on installing basic gravity sensing devices on vehicles. Sensors on the vehicle collect overall, individual weight or pressure data. Operators store and analyze the massive amounts of collected data offline in the cloud, attempting to filter out potential overloading behavior records.

[0006] While this scheme represents an effort by operators in technical oversight, it has revealed three fundamental flaws in practical application, leading to its poor effectiveness and failure to gain widespread adoption: Flaw one: Severe latency and a lack of real-time interference capabilities are the most fatal flaws, causing architectural delays. The entire "collection-upload-analysis" process results in slow data processing and extremely untimely feedback, meaning the system cannot detect and prevent overload events instantly. This "post-event traceability" model, which relies on a data recording tool and a proactive security intervention system, cannot eliminate the danger at its root.

[0007] Defect 2: The judgment logic is simple, with low accuracy and susceptibility to interference. Relying solely on gravity sensor data, the judgment logic is very crude and fragile. It can only sense "there is a button," but cannot understand "what it is." This leads to many incorrect judgment scenarios: ① Inability to distinguish between people and objects: A lighter cyclist carrying a heavy backpack may have a total weight similar to two lighter cyclists. Simple gravity sensing cannot distinguish between these two completely different situations. ② Inability to recognize complex scenarios: A 90 kg cyclist and "a 60 kg cyclist carrying a 30 kg child" may appear indistinguishable to a simple gravity sensor, but the unprecedented safety risks go far beyond this. ③ Difficulty in setting thresholds: Setting the weight threshold too low will misjudge many normal users carrying heavy objects; setting the threshold too high will miss a large number of violations.

[0008] Defect 3: Poor user experience. Due to the aforementioned delays and inaccuracies leading to technical failures, the actual user experience of this technical solution is extremely poor. Users cannot obtain timely and clear feedback, and operators find it difficult to implement effective incentive measures due to low penalty rates. Ultimately, this technology is restricted in its use, resulting in overloading rates as high as 35% in areas like Nanchang, despite signs indicating "One Person Only," rendering regulations largely ineffective.

[0009] Although there have been some initial attempts in the industry, the commonly used method is a perceptual, lagging, single-dimensional (total weight) monitoring and analysis method. This method has fundamental architectural and logical flaws that fail to meet the needs of effectively managing overload behavior in terms of real-time performance, accuracy, and user experience. Summary of the Invention

[0010] The purpose of this invention is to address the shortcomings of the prior art by providing a method, system, device, and medium for intelligent identification of overloaded shared bicycle riders, thereby solving the problems in the prior art.

[0011] This invention specifically provides the following technical solution: a method for intelligent identification of overloaded shared bicycle riding, comprising the following steps: collecting the temperature and pressure of the shared bicycle seat through a sensor array inside the shared bicycle seat; reshaping multiple discrete temperatures and pressures into two independent matrices to obtain an original pressure map and an original temperature map; constructing a three-channel input tensor through the original pressure map and the original temperature map; the three-channel input tensor includes: a normalized pressure map, a pressure gradient map obtained based on the normalized pressure map, and a dynamic temperature difference map obtained based on the original temperature map and a dynamic environmental reference temperature, wherein the dynamic environmental reference temperature is predicted based on historical data; performing pattern recognition on multiple input tensors to obtain classification results belonging to different riding scenarios, and controlling the shared bicycle to perform corresponding operations according to the classification results, wherein when the classification result is an overloaded scenario, an alarm is triggered or the vehicle operation is restricted.

[0012] Preferably, the dynamic temperature difference map obtained based on the original temperature map and the dynamic environmental reference temperature is as follows: when the user unlocks the vehicle, the initial environmental temperature of the shared bicycle seat is collected and uploaded to the cloud; the cloud model combines time, geographical location information and historical thermodynamic data of the vehicle to predict a dynamic environmental reference temperature curve showing the temperature change of the shared bicycle seat over time when it is unoccupied; during real-time detection, the difference between the temperature value of each point in the original temperature map and the dynamic environmental reference temperature curve at the corresponding time is obtained to generate the dynamic temperature difference map.

[0013] Preferably, the step of performing pattern recognition on multiple input tensors to obtain classification results belonging to different cycling scenarios specifically involves: inputting each input tensor into the convolutional layer of a convolutional neural network model, and passing it through a transistor of size [size missing]. convolution kernel The image is slid across the image to perform a dot product operation, resulting in an output feature map. The output feature map is then max-pooled and flattened into a one-dimensional vector. This one-dimensional vector is then mapped to a space with dimensions below a threshold through a fully connected layer. The output of the fully connected layer is converted into a probability distribution vector, and the classification results for different cycling scenarios are obtained through this probability distribution vector.

[0014] Preferably, the convolutional neural network model is trained using sample data containing multiple cycling scenarios, which include at least: an empty state, a single adult cycling state, an overloaded state with an adult carrying a child, an overloaded state with two adults, a state with an adult carrying an object, and a state with only inanimate objects placed on the ground.

[0015] Preferably, the convolutional neural network model is constructed by fusing multi-scale algorithms, asymmetric convolutional kernels, and residual network structures, and is trained using the Adam optimization algorithm and a learning rate decay strategy.

[0016] Preferably, when collecting the temperature and pressure of the shared bicycle seat, the specific method is as follows: the digital signal output by the sensor is converted into degrees Celsius, and the specific expression is: ;in, In the sensor array The actual temperature value at the location, This is the raw digital reading read from the position sensor; the collected resistance change is converted into a pressure value, specifically expressed as: ;in, Is The pressure value at the location, It is the measured change in resistance. It is the sensor sensitivity coefficient that needs to be calibrated through experiments.

[0017] Preferably, the pressure sensor is an IMM-00039 series distributed multi-point pressure sensing resistive sensor, and the temperature sensor is a DS18B20 temperature sensor.

[0018] This invention provides an intelligent overload detection system for shared bicycles, comprising: a data acquisition module for collecting temperature and pressure data of the shared bicycle seat using a sensor array within the seat; an image conversion module for reshaping multiple discrete temperature and pressure data into two independent matrices to obtain an original pressure map and an original temperature map; a three-channel module for constructing a three-channel input tensor using the original pressure map and the original temperature map; the three-channel input tensor includes: a normalized pressure map, a pressure gradient map obtained based on the normalized pressure map, and a dynamic temperature difference map obtained based on the original temperature map and a dynamic environmental reference temperature, wherein the dynamic environmental reference temperature is predicted based on historical data; and a detection module for performing pattern recognition on the multiple input tensors to obtain classification results belonging to different riding scenarios, and controlling the shared bicycle to perform corresponding operations based on the classification results, wherein when the classification result is an overload scenario, an alarm is triggered or the vehicle operation is restricted.

[0019] The present invention provides a computer device, including a memory and a processor. The memory stores a program, and when the program is executed by the processor, the processor performs the steps of the above-described intelligent identification method for overloaded shared bicycles.

[0020] The present invention provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described intelligent identification method for overloading of shared bicycles.

[0021] Compared with existing technologies, the present invention has the following significant advantages: The present invention collects the temperature and pressure of shared bicycle seats, reshapes them to obtain original pressure and temperature maps, and constructs a three-channel input tensor using the original pressure and temperature maps. Pattern recognition is then performed on the input tensor to obtain classification results for different riding scenarios. The present invention upgrades a simple overload detection problem into a data-driven image classification task, treating the collected sensor data points as a single information-rich multi-channel "pressure image." Each channel provides a unique dimension for subsequent intelligent analysis, thereby improving the accuracy of subsequent recognition. Simultaneously, by recognizing each channel's tensor, the overload judgment result (such as "overloaded," "normal," or "carrying only cargo") is fed back to the vehicle to perform corresponding operations (such as alarm or power cut-off), thereby achieving a real-time, accurate, reliable, and intelligent overload management effect. Attached Figure Description

[0022] Figure 1 is a flowchart of the judgment logic in this invention; Figure 2 is a block diagram of the system structure in this invention; Figure 3 is a planar design diagram of the shared electric vehicle seat and shared bicycle seat cushion in this invention; Figure 4 is a 3D model design diagram of the shared electric vehicle seat and shared bicycle seat cushion in this invention; Figure 5 is a schematic diagram of the one-dimensional convolutional neural network model structure in this invention; Figure 6 is a temperature measurement and prediction curve diagram under the condition of bicycle unlocking in spring in this invention; Figure 7 is a temperature measurement and prediction curve diagram under the condition of bicycle unlocking in summer in this invention; Figure 8 is a temperature measurement and prediction curve diagram under the condition of bicycle unlocking in autumn in this invention. Figure 9 shows the temperature measurement and prediction curves under the condition of unlocking a single bicycle in winter according to the present invention; Figure 10 shows the feature diagram under the idle state according to the present invention; wherein Figure 10(a) is feature diagram 1: pressure diagram, Figure 10(b) is feature diagram 2: pressure gradient diagram; Figure 10(c) is feature diagram 3: temperature difference diagram (°C), environmental reference: 34.5°C; Figure 11 shows the feature diagram under the condition of a single adult according to the present invention; wherein Figure 11(a) is feature diagram 1: pressure diagram, Figure 11(b) is feature diagram 2: pressure gradient diagram; Figure 11(c) is feature diagram 3: temperature difference diagram (°C). Figure 12 shows the characteristic diagram of an adult with a child in this invention; wherein Figure 12(a) is characteristic diagram 1: pressure diagram, Figure 12(b) is characteristic diagram 2: pressure gradient diagram; Figure 12(c) is characteristic diagram 3: temperature difference diagram (°C), environmental reference: 15.2°C; Figure 13 shows the characteristic diagram of two adults in this invention; wherein Figure 13(a) is characteristic diagram 1: pressure diagram, Figure 13(b) is characteristic diagram 2: pressure gradient diagram; Figure 13(c) is characteristic diagram 3: temperature difference diagram (°C), environmental reference: 15.2°C; Reference temperature: 33.6°C; Figure 14 is a feature diagram of the adult + object state in this invention; wherein Figure 14(a) is feature diagram 1: pressure diagram, Figure 14(b) is feature diagram 2: pressure gradient diagram; Figure 14(c) is feature diagram 3: temperature difference diagram (°C), environmental reference temperature: 29.0°C; Figure 15 is a feature diagram of the object-only state in this invention; wherein Figure 15(a) is feature diagram 1: pressure diagram, Figure 15(b) is feature diagram 2: pressure gradient diagram; Figure 15(c) is feature diagram 3: temperature difference diagram (°C), environmental reference temperature: 28.8°C. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0024] The core objective of this invention is to fundamentally solve the widespread overloading problem of shared electric bicycles through technical means, ultimately achieving the ideal state of "one person, one bike," thereby bringing a series of positive and inevitable benefits. Specifically, achieving the desired effects can be summarized in the following three aspects:

[0025] Significantly improving traffic safety and eliminating safety hazards: The most direct and important goal is to ensure cycling safety. A crucial intelligent system capable of detecting and proactively triggering overloading in real time can effectively prevent the dangerous behavior of multiple riders. This not only protects the cyclists themselves but also reduces the risks of vehicle loss of control and brake failure caused by overloading, thereby greatly reducing potential traffic safety hazards and contributing to urban traffic safety.

[0026] Protecting Shared Assets and Extending Vehicle Lifespan: From the operator's perspective, the purpose of this invention is to protect their core assets. Overloading inevitably increases the accelerated wear and tear on the vehicle frame, motor, battery, and tires, significantly reducing the overall lifespan of the shared electric bicycle. By effectively preventing overloading, it is possible to ensure that the vehicle operates within the designed system load-bearing capacity, thereby extending the lifespan of the shared electric bicycle and reducing the operator's depreciation and maintenance costs.

[0027] Providing an efficient management solution for effective supervision: This system aims to offer shared electric bicycle operators a concrete, convenient, and efficient management solution, addressing the current limitations of "paper regulations" and "post-event supplementation" management models. Through closed-loop management with real-time detection and timely feedback (such as alarms and power outages), this system ensures that regulations are effectively implemented, enabling operators to manage their vehicles more effectively and sustainably. This not only improves operational efficiency but also helps companies better fulfill their social safety responsibilities, promoting the standardized development of the entire industry.

[0028] In summary, the ultimate goal of this invention is to create an intelligent detection and feedback system that integrates safety, asset protection, and efficient management, thereby completely changing the current chaotic situation of overloading of shared electric bicycles and achieving safe, standardized, and sustainable operation.

[0029] To achieve accurate, real-time, and precise intelligent identification of overloading behavior in shared electric bicycles, this technical solution abandons the simple logic of traditional fixed thresholds and adopts a machine learning method based on "pressure image" recognition and dynamic environment adaptive algorithms. This solution fundamentally innovates the data processing and decision-making methods based on dual-sensor hardware (temperature and pressure).

[0030] The specific implementation method consists of the following three core parts: 1. Sensor data visualization: Constructing a multi-channel "pressure image": The foundation of this solution is a high-density sensor array installed inside the gasket, containing multiple pressure sensors (IMM-00039 series) and temperature sensors (DS18B20 series) with a total size of 52×44 (2288 in total). The key innovation lies in not isolating these 2288 discrete data points, but rather treating them as a unified, information-rich multi-channel "pressure image." This "image" consists of three independent feature channels, each providing a unique dimension for subsequent intelligent analysis:

[0031] Channel 1: Normalized Pressure Map: After collecting and normalizing the raw readings of 2288 pressure sensors, a grayscale map is generated that captures the distribution, magnitude, and basic shape of the pressure.

[0032] Channel 2: Pressure Steepness Map: This map generates a steepness map that highlights objects and edges by calculating the rate of change between boundary pressure points. This map is crucial for distinguishing between the human body (smooth) and hard objects (like boxes, which are equally sharp).

[0033] Channel 3: Dynamic Temperature Difference Map: This is the core of our response to environmental changes. This channel does not display absolute temperature, but rather the difference between the actual measured temperature and the "ambient baseline temperature" predicted by the cloud model. It effectively filters out environmental interference (such as summer sun exposure) and highlights only "abnormal heat sources" generated by living organisms.

[0034] 2. Image Pattern Recognition: Intelligent classification using neural networks (CNN): After converting sensor data into images, neural networks (CNN) are used to replace rule-based microcontroller judgment logic.

[0035] Model training: Deploy a customized lightweight CNN model in the cloud and train it using thousands of sets of "stress images" covering various real-world scenarios (such as a single adult, an adult and a child, an adult and a backpack, objects placed on the back, etc.).

[0036] Intelligent decision-making: A well-trained CNN model can automatically learn and recognize complex spatial patterns in different "stress images," rather than simply calculating the number of hotspots or comparing total stress levels. For example, the model can:

[0037] By the shape and number of pressure zones, we can distinguish between "one cyclist" (single, large pressure zone) and "two cyclists" (two overlapping or side-by-side large pressure zones); by the strictness of the pressure zones, we can distinguish between "child" (mild) and "backpack" (both sharp); by the number of heat source signals (from the dynamic temperature difference map), we can accurately distinguish between "a 60 kg cyclist carrying a 30 kg child" (two pressure zones, two heat sources) and "a 60 kg cyclist carrying a 30 kg backpack" (two pressure zones, but only one heat source).

[0038] 3. Dynamic Environment Planning: Achieving Adaptability: In order to solve the fundamental problem that traditional temperature thresholds fail under extreme weather conditions, this solution introduces a dynamic environment adjustment algorithm.

[0039] Establishing a dynamic baseline: When a user unlocks a shared bicycle, the system immediately collects the initial ambient temperature of the bicycle seat and uploads it to the cloud. The cloud model combines the current time, time zones (which can be correlated with weather data), and historical thermodynamic data of the vehicle to predict a "preset baseline curve" showing how the shared bicycle seat temperature changes over time when the bicycle is unoccupied. The "preset baseline curves" for spring, summer, autumn, and winter are shown in Figures 6, 7, 8, and 9, respectively.

[0040] Calculate the relative temperature difference: In subsequent real-time detection, the system uses the difference between the actual measured temperature and the dynamic reference, rather than the absolute temperature value, to make a judgment.

[0041] Achieving robustness: A person sitting on a shared bicycle seat exposed to the sun at 45°C will still have their body temperature affect the evaporation rate in that area, resulting in a significant positive temperature difference. This positive temperature difference is even more pronounced in cold winters. In this way, the human body heat source signal becomes a stable and identifiable feature under any weather conditions, overcoming the drawbacks of the static thresholding method.

[0042] In summary, this technical solution transforms a simple overload detection problem into a complex, data-driven image classification task through three core methods: data sensor imaging, CNN intelligent recognition, and dynamic environment planning. The onboard device (C8051F341 microcontroller) efficiently collects and transmits data, while all complex intelligent analysis and decision-making are completed in the cloud. Ultimately, the overload judgment result (e.g., "overloaded," "normal," "carrying only cargo") is fed back to the vehicle to execute corresponding actions (e.g., alarm, power cut-off), thereby achieving accurate, reliable, and intelligent overload control.

[0043] Based on the above description, as shown in Figure 1, this embodiment provides a method for intelligent identification of overloaded shared bicycles, specifically including: Step S1: Collecting the temperature and pressure of the shared bicycle seat through a sensor array set inside the shared bicycle seat.

[0044] Raw data conversion: When collecting temperature and pressure data from shared bicycle seats, specifically: Temperature data conversion: The DS18B20 sensor outputs a digital signal, which needs to be converted to degrees Celsius (°C). The specific expression for converting the sensor's digital output signal to degrees Celsius is as follows: ;in, In the sensor array The actual temperature value at the location, This is the raw digital reading from the position sensor.

[0045] Pressure Data Conversion: The IMM-00039 series resistive sensor outputs a change in resistance. This needs to be converted into a pressure value (e.g., kilopascals, kPa). This conversion depends on the sensor's sensitivity constant. The collected resistance change is converted into a pressure value, and the specific expression is as follows:

[0046] ;in, Is The pressure value at the location, It is the measured change in resistance. It is the sensor sensitivity coefficient that needs to be calibrated through experiments.

[0047] The sensor array includes a pressure sensor and a temperature sensor. The pressure sensor is an IMM-00039 series distributed multi-point pressure sensing resistive sensor, and the temperature sensor is a DS18B20 temperature sensor.

[0048] The DS18B20 temperature sensor is a commonly used temperature sensor with a wide temperature measurement range and high resolution, meeting the temperature measurement needs in most situations. It can also directly output the measured temperature as a binary number, which can be controlled by a microcontroller to display the analysis results on an LED screen. This solution determines the number of heat sources by comparing the number of temperature peaks detected by passengers (contact sources) and then transmitting the data to the microcontroller for analysis. Typical application scenarios and values ​​are shown in Table 1, and a comparison with alternative solutions is shown in Table 2. It also has the following advantages:

[0049] 1. Wide temperature range and high accuracy meet requirements: The DS18B20's operating temperature range (-55°C to +125°C) fully covers the extreme environments that shared bicycles may encounter (such as the surface temperature of shared bicycle seats can reach above 70°C under summer sun); the accuracy of ±0.5°C is sufficient to monitor the temperature changes of shared bicycle seats and to provide early warning of the risk of high-temperature burns or low-temperature icing.

[0050] 2. Simplified wiring with a single bus: Only one data cable (+ power / ground) is needed for communication, which greatly reduces the complexity of the cable from the shared bicycle seat to the main control of the lock, making it suitable for the lightweight and reliability requirements of shared bicycles.

[0051] 3. Low power consumption and power supply compatibility: The static current is only 1μA and the operating current is about 1mA. It is compatible with the solar + lithium battery power supply solution of shared bicycle smart locks (such as 5V system) and has minimal impact on the overall power consumption.

[0052] Key Challenges and Solutions in the Shared Bike Scenarios: 1. Enhanced Environmental Tolerance: Waterproof and Dustproof: The original DS18B20 package offers no protection; an IP68-level seal (such as silicone potting or heat shrink tubing) is required to withstand rain immersion and dust corrosion. Mechanical Protection: Frequent pressure on shared bike seats may damage sensors. It is recommended to embed sensors inside the shared bike seat rather than on the surface, and use an elastic cushioning layer (such as EVA foam) to distribute pressure.

[0053] 2. Temperature Measurement Accuracy Guarantee: Thermal Coupling Design: The sensor must be tightly fitted to the heat-conducting layer of the shared bicycle seat (such as a metal frame) to avoid air gaps causing temperature measurement lag. Isolation of External Interference: Adding a heat insulation layer (such as ceramic fiber) blocks the influence of direct sunlight or heat conduction from the frame, ensuring that the temperature reflects the true body temperature.

[0054] 3. System Integration Optimization: Interface Expansion: The main controller of the shared bicycle smart lock (such as STM32F103) usually reserves GPIO, which can be expanded to DS18B20 through a single bus without the need for an additional communication module. Power Consumption Coordination: An intermittent sampling strategy (such as waking up once every 10 minutes) is adopted, combined with the smart lock's sleep mode, to further reduce power consumption.

[0055] Implementation Recommendations: 1. Hardware Design: The sensor is embedded in the bottom layer of the shared bicycle seat and connected to the main control of the lock via an FPC flexible cable to avoid bending damage; the power supply is taken from the 5V voltage regulator module of the smart lock and connected in series with a self-resetting fuse to prevent short circuit.

[0056] 2. Operation and maintenance strategy: Combine vehicle scheduling and maintenance cycle (1~2 weeks) to conduct spot checks on sensor sealing and data accuracy; establish a temperature history database in the cloud and train an abnormal temperature prediction model (such as prioritizing vehicle scheduling in areas exposed to direct sunlight).

[0057] The DS18B20 offers an excellent balance of cost, accuracy, and integration, making it a practical choice for monitoring the temperature of shared bicycle seats. However, environmental adaptability issues must be addressed through waterproofing, reinforcement, mechanical protection, and thermal design.

[0058] The IMM-00039 series distributed multi-point pressure-sensing resistive sensor deforms under external pressure, altering the contact area between materials and thus changing the resistance. The peak pressure signal generated by a rider (pressure source) sitting on a shared electric scooter is compared, and the collected data is transmitted to a microcontroller for analysis. A comparison with existing solutions is shown in Table 3.

[0059] Adaptation advantages: 1. Distributed multi-point measurement capability: This series of sensors adopts a multi-point layout, which can cover the pressure distribution of different areas of the shared bicycle seat, accurately detect the changes in the rider's center of gravity, and is suitable for scenarios such as multi-person riding recognition and riding posture analysis; compared with single-point sensors, the multi-point design can reduce misjudgments (such as distinguishing between the placement of objects and the weight of the human body).

[0060] 2. Flexible structure and easy integration: Resistive sensors typically have a flexible film or fabric substrate that can fit the curved surface of shared bicycle seats without the need for complex mechanical modifications; the thin design (typical thickness <1mm) does not affect the comfort of shared bicycle seats and is easy to embed into existing shared bicycle seat structures.

[0061] 3. Cost and power consumption advantages: The cost of resistive sensors is significantly lower than that of piezoresistive MEMS solutions (such as Honeywell SSC series), making them suitable for large-scale deployment; the power consumption is extremely low during static detection (power is only required when sampling), which meets the battery life requirements of shared bicycles.

[0062] Potential challenges and improvement needs: 1. Environmental tolerance: Waterproof and dustproof: Shared bicycles need to meet at least IP67 protection level, while resistive sensors are susceptible to humid environments and require additional encapsulation protection.

[0063] 2. Mechanical durability: It must withstand long-term sitting pressure and vibration (reference standard: >100g acceleration impact 1) to avoid fatigue fracture of the resistive layer.

[0064] Optimization suggestions: 1. System integration solution: Adopt a time-division sampling strategy to reduce power consumption and activate the sensor only during vehicle use.

[0065] 2. Environmentally Adaptable Design: An additional silicone sealing layer is added to achieve IP68 protection while maintaining pressure sensitivity; a cushioning structure is designed at the bottom of the shared bicycle seat to distribute impact loads.

[0066] 3. Algorithm Collaboration: By combining algorithms, when the pressure sensor detects a change in load, it helps to determine the vehicle status.

[0067] Step S2: Reshape multiple discrete temperatures and pressures into two independent matrices to obtain the original pressure map and the original temperature map.

[0068] A microcontroller is used as the main control center to control various circuits to complete data acquisition and transmission.

[0069] Image matrix construction: 2288 discrete pressure readings and temperature readings Remodeled into two independent The matrix, which this invention refers to as the original pressure map. and the original temperature map .

[0070] Mini-Max Normalization: To eliminate dimensional differences and make model training more stable, all data needs to be normalized to intervals. Apply the following formula to each pixel in the stress map:

[0071] ;in, and These are the minimum and maximum pressure readings obtained within a specific time window or based on statistics from a large number of samples. Temperature graphs are processed similarly.

[0072] Step S3: Construct a three-channel input tensor using the original pressure map and the original temperature map; the three-channel input tensor includes: a normalized pressure map, a pressure gradient map calculated based on the normalized pressure map, and a dynamic temperature difference map calculated based on the original temperature map and the dynamic environmental reference temperature.

[0073] To provide the model with richer information, this design constructs a three-channel input tensor. .

[0074] Channel 1: Normalized Pressure Chart The specific expression is: Channel 2: Pressure Gradient Plot This channel highlights the edges and contours of the pressure. A simple finite difference method is used to approximate the gradient for each point in the image. Its gradient magnitude can be calculated as:

[0075] Among them, the horizontal gradient and vertical gradient They are respectively: (For image boundaries, forward or backward differencing can be used.) The calculated gradient map G, after normalization, becomes the second channel. .

[0076] Channel 3: Dynamic Temperature Difference Chart This is the core of achieving dynamic adaptive thresholding; the dynamic temperature difference map, based on the original temperature map and the dynamic environmental reference temperature, specifically involves: when a user unlocks the vehicle, the initial environmental temperature of the shared bicycle seat is collected and uploaded to the cloud; the cloud model, combining time, geographical location information, and historical thermodynamic data of the vehicle, predicts a dynamic environmental reference temperature curve showing the temperature change of the shared bicycle seat over time when the bicycle is unoccupied; during real-time detection, the value of this channel is not an absolute temperature, but rather the difference between the temperature value at each point in the original temperature map and the corresponding dynamic environmental reference temperature curve, generating the dynamic temperature difference map; the specific expression is: ;in, It is a dynamic environment calibration model in time Position The predicted baseline temperature. This difference plot (which also needs to be normalized) highlights the "abnormal heat sources" generated by the human body, effectively eliminating interference from ambient temperature.

[0077] Step S4: Perform pattern recognition on multiple input tensors to obtain classification results belonging to different cycling scenarios, and control the shared bicycles to perform corresponding operations based on the classification results. When the classification result is an overload scenario, trigger an alarm or restrict vehicle operation.

[0078] Pattern recognition employs a Convolutional Neural Network (CNN) model: Multiple input tensors are processed for pattern recognition to obtain classification results belonging to different cycling scenarios. Specifically: 1. Each input tensor is input into the convolutional layer of the CNN model, i.e., for one channel of the input image. Through a size of Convolution kernel (or filter) Slide the image to perform a dot product operation and obtain the output feature map.

[0079] Among them, the convolutional layer (Conditional Layer): Convolution is the core of the model for feature extraction. Output feature map. an element on The calculation formula is:

[0080] ;in, Is it the convolution kernel in The location of the position, It is a bias term. It is an activation function that introduces non-linearity, enabling the model to learn more complex patterns.

[0081] 2. Max pool the output feature map to flatten it into a one-dimensional vector, and then use a fully connected layer to map the one-dimensional vector to a space with a dimension below the threshold. The output of the fully connected layer is converted into a probability distribution vector, and the classification results belonging to different cycling scenarios are obtained through the probability distribution vector.

[0082] Among them, the pooling layer (Max Pooling) is used to reduce the spatial dimensionality of the data, reduce computation, and provide translation invariance of features. It divides the feature map into several non-overlapping rectangular regions (e.g., 2*2) and outputs the maximum value in each region.

[0083] Among them, the fully connected layer: after several rounds of convolution and pooling, the final feature map is flattened into a one-dimensional vector. Fully connected layers map this to a lower-dimensional space. The operation is essentially a matrix multiplication followed by an activation function:

[0084] ;in, It is a weight matrix; It is the bias vector; It is the output vector of this layer.

[0085] The output layer and the softmax function: The last layer is the output layer, and the number of neurons in it is equal to the total number of categories. (For example, (Number of categories). It uses the Softmax function to optimize the output of the fully connected layer. Transform into a probability distribution vector :

[0086] ;in, The model predicts the category to which the input image belongs. The probability of all The sum is 1. Figure 5 is a schematic diagram of the structure of a one-dimensional convolutional neural network model.

[0087] The convolutional neural network model is trained using sample data containing various cycling scenarios, which include at least: an empty state, a single adult riding, an overloaded state with an adult carrying a child, an overloaded state with two adults, a state with an adult carrying an object, and a state with only inanimate objects placed on the ground.

[0088] The core of this invention lies in its intelligent scene classification capability. It no longer relies on any single, fixed temperature or pressure threshold, but instead makes accurate judgments by comprehensively analyzing the complete "pressure image" collected by the sensor array. Like an experienced analyst, the system can interpret the magnitude, shape, contour, and crucial heat source signals of the pressure, thereby identifying six core cycling scenarios.

[0089] Idle State: The system can identify completely empty shared bicycle seats. In this case, the "pressure image" only shows low-intensity background noise with no fixed shape, and its temperature is completely consistent with the dynamic environmental benchmark provided by the cloud model, without any abnormal heat source signals.

[0090] Single adult: When an adult is riding normally, the system will identify a large, coherent pressure area, accompanied by a distinct, strong heat source signal that perfectly matches it. This is the baseline pattern for judging compliant riding.

[0091] Overload - Adults and Children: This is a key manifestation of the system's intelligence. In this scenario, the "pressure image" will present two pressure zones of different sizes (one for adults and one for children). The most crucial identification criterion is that the system will simultaneously detect two independent heat source signals, corresponding to these two pressure zones respectively.

[0092] Overload - Two adults: When two adults are riding, the pressure zone covers most of the shared bicycle seat, is extremely irregular in shape, and is accompanied by a wide range of high-intensity heat source signals. Two pressure and temperature peaks can usually be identified.

[0093] Carrying Objects - Adults and Backpacks: To resolve the confusion between "children" and "backpacks," the system employs a unique recognition logic for such scenarios. The image displays two pressure zones, but the system only detects the heat source signal matching the adult's pressure zone. The other pressure zone, belonging to an object, regardless of how hot it is in summer, will not be identified as a heat source because its temperature is essentially the same as the environmental baseline.

[0094] Object-only - No living objects only: To eliminate false alarms caused by placing heavy objects, the system can accurately identify scenarios where only objects are placed. In this case, although a clear pressure area will be displayed on the "pressure image" (usually with a sharper outline than a human body), the system will not detect any effective heat source signal, thus determining it to be a non-human-occupied state.

[0095] This system achieves a deep understanding of the nature of riding scenarios by intelligently correlating and analyzing the spatial distribution of pressure with heat source signals, rather than simply measuring weight. This capability enables it to accurately distinguish between compliant riding, different types of overloading, and cargo carrying in various complex and extreme environments, thereby achieving efficient and reliable supervision. The specific classifications are shown in Table 4 below.

[0096] The convolutional neural network model is constructed using a fusion of multi-scale algorithms, asymmetric convolutional kernels, and residual network structures, and trained using the Adam optimization algorithm and a learning rate decay strategy. To train the model, a loss function is needed to measure the difference between the predictions and the true labels. For multi-class classification problems, the classification cross-entropy loss function is used:

[0097] ;in, It is a one-not encoded vector of the true label (if the true category is...). ,but Otherwise ); The model predicts that it belongs to a category. The probability. The goal of training is to adjust all the weights (K and W) and biases (b and W) in the network. To minimize this loss function .

[0098] Once the model is trained, the final overload judgment calculation for a new input stress image is very simple: 1. The data is processed through the above preprocessing and feature engineering steps to generate a three-channel input tensor. 2. Tensors 3. Input the data into the trained model. The model then performs forward propagation, ultimately outputting a probability vector. 4. The final classification result is determined by the category with the highest probability:

[0099] Predicted Class= .

[0100] Table 1 Typical Application Scenarios and Value Table 2 Comparison of Alternative Solutions Table 3 Comparison with existing solutions In terms of sensor arrangement, this solution uses 52 rows × 44 columns of pressure sensors and temperature sensors interspersed on the seat cushion of the Qingju electric shared bicycle, which is 40.5cm long and 30.5cm wide, to collect a total of 2288 pressure sensors, which can maximize the collection of rider data.

[0101] As shown in Figure 2, the C8051F341 microcontroller is used as the main control center to control various circuits for data acquisition and transmission. Sensors convert pressure and temperature signals into electrical signals. Since the system uses multiple sensors for data acquisition, a multiplexer selects the data. The selected data is amplified by a data amplifier and then converted from analog to digital signals by an A / D converter. This digital signal is then sent to the microcontroller for processing, and the processed data is used to alert the user. If the cyclist is overloaded before riding, the microcontroller will issue a warning via a buzzer and status indicator light and cut off power until the cyclist stops overloading. During riding, the system monitors in real time. If the cyclist is overloaded, the system will sound an alarm to remind the cyclist to ride properly, thus reducing overloading. A reset circuit and an oscillation circuit are also included.

[0102] Core Advantages and Adaptability: 1. High-performance mixed-signal processing capability: Integrated 8-channel 12-bit ADC (100ksps sampling rate), which can be directly connected to analog signal sources such as pressure sensors, temperature sensors, and Hall sensors, eliminating the need for external ADC chips and simplifying circuit design; Built-in 2-channel UART and SPI interfaces, which can simultaneously connect to a GPS module (UART1) and a GPRS / 4G communication module (UART2), supporting remote data transmission; 25MIPS high-speed 8051 core, meeting the computing power requirements for real-time processing of multi-sensor data (such as pressure analysis of shared bicycle seat cushions and positioning data encapsulation) and control logic (such as motor lock control).

[0103] 2. Low power consumption and power management: Supports multiple sleep modes (idle / stop mode), with a sleep current as low as 0.1μA. It can be woken up by interruption (such as vibration sensor trigger), which can significantly extend the battery life of a single vehicle; Independent power monitoring module (SMBus) can monitor battery voltage in real time, trigger low battery warning and upload to the cloud.

[0104] 3. Scalability and integration: 64KB Flash + 4.25KB RAM, sufficient to store map cache, user riding logs and OTA firmware upgrades; the digital crossbar can flexibly map peripherals (PWM, timers) to I / O pins to adapt to different hardware layouts.

[0105] Key challenges and solutions in the shared bicycle scenario: 1. Enhanced environmental tolerance: Waterproof and dustproof: requires encapsulation in an IP67-rated protective box and coating the PCB with conformal coating; Wide temperature operation: industrial-grade temperature range (-40°C to +85°C) requires the use of a temperature sensor (such as DS18B20) and dynamic power consumption adjustment algorithm, high temperature frequency reduction and low temperature heating.

[0106] 2. Multi-sensor collaborative acquisition: Pressure sensing: IMM-00039 resistive sensor array is connected via ADC, and multi-point pressure fusion algorithm is implemented in software (to identify two riders); Positioning and attitude: 6-axis IMU (such as MPU6050) is connected via SPI interface, and GPS data is used to compensate for positioning drift; Motor control: PWM output drives brushless motor (bicycle lock), and Hall sensor feedback of position forms a closed loop.

[0107] 3. Remote communication reliability: Dual-mode transmission design: Main channel: UART connects to the Quectel BG96 4G module, uploading data to the cloud platform via AT commands; Backup channel: SPI connects to the LoRa module (such as SX1276), enabling Mesh self-organizing network transmission in signal dead zones; Data compression optimization: The TinyFLAT protocol is used to compress positioning / sensor data packets, reducing data consumption.

[0108] After combining the microcontroller with two sensors, the following shared bicycle seat model was designed: 1- Temperature sensor: When measuring temperature, the microcontroller and DS18B20 temperature sensor need to be connected; 2- Pressure sensor: IMM-00039 series distributed multi-point pressure sensing resistive sensor; 3- Microcontroller signal receiver: The overall system design uses the C8051F341 microcontroller as the main control center to control various circuits to complete data acquisition and transmission; 4- Shared bicycle seat style is shown in Figures 3 and 4: This scheme takes the Qingju shared electric bicycle seat as an example to simulate the deformation of the shared bicycle seat under the force of whether or not there is a rider.

[0109] Table 4. Cycling Scenarios Classification The implementation of this technical solution represents a fundamental shift from a fragile, rule-based detection system to an intelligent, adaptive, and high-precision recognition system. Its technical effects are mainly reflected in the following four aspects:

[0110] 1. Fundamental Improvement in Accuracy and Reliability: The core technological improvement is a significant increase in the accuracy and reliability of overload detection. Traditional solutions rely solely on calculating the number of temperature peaks and comparing a single pressure threshold, making them prone to false positives. This solution treats sensor data as a "pressure image" and applies a convolutional neural network (CNN) model for analysis, achieving a deep understanding of the riding scenario. Results: The system can accurately distinguish pressure patterns generated by different objects (such as human bodies, backpacks, and boxes) because it analyzes the shape, contour, centroid, and gradient of the entire pressure distribution, not just the weight value. This effectively curbs common false alarms (e.g., misjudging a heavy backpack as overloaded) and false negatives (e.g., two lighter cyclists riding together but their total weight is within the limit).

[0111] 2. Stable operation in all weather and environments: This solution successfully addresses the critical issue of traditional temperature sensors failing in extreme weather conditions, achieving all-weather robustness. Demonstration: By introducing a "dynamic environmental calibration" algorithm, the system no longer relies on a fixed temperature threshold (e.g., 32°C). Whether in the scorching heat of summer reaching 45°C or the frigid winter at 5°C, the system can reliably identify "abnormal heat sources" in the human body by calculating the difference between the actual temperature and the dynamic environmental benchmark. This ensures that the system's judgment logic is unaffected by ambient temperature, maintaining efficient and consistent performance under various climatic conditions.

[0112] 3. Achieving Refined and Intelligent Scene Classification: Another major leap in technological advancement is that the system no longer simply makes binary judgments of "overloaded / not overloaded," but can achieve refined classification of riding scenarios. The effect is manifested in the system's ability to clearly distinguish between complex scenarios with completely different characteristics. For example, it can accurately identify the essential difference between "a 60kg cyclist carrying a 30kg child" (image features: two pressure areas, matched with two independent heat sources) and "a 60kg cyclist carrying a 30kg backpack" (image features: two pressure areas, but only one heat source). This capability elevates supervision from simple weight detection to intelligent identification of the nature of riding behavior, providing a solid technological foundation for tiered response (e.g., only prompting for cargo, cutting off power for passengers).

[0113] 4. Real-time and proactive safety intervention capabilities: Compared to the slow offline analysis model of some operators who wait for data to be uploaded, this solution achieves near real-time feedback and proactive intervention. The entire process, from data collection at the vehicle end to intelligent analysis and command return (such as issuing alarms or cutting off power) in the cloud, is optimized to a second-level response time. This allows the system to intervene immediately before or during unsafe riding behavior, preventing dangerous overloading at its source, rather than relying on post-incident intervention. This not only significantly improves traffic safety but also effectively protects shared electric bicycles and extends their lifespan.

[0114] The practical effect of this technical solution is the successful creation of a high-precision, highly reliable, environmentally unaffected, and deeply understanding passenger scenarios intelligent monitoring system. It fundamentally solves the limitations of traditional solutions, elevating overload detection to a completely new level of intelligence.

[0115] Experimental data: Data from relevant experiments support the above effects. (For effects with multiple characteristics, it is best to use multiple experiments to characterize them separately):

[0116] The process of data acquisition and transmission by temperature and gravity sensors was simulated using Proteus software to observe and evaluate the system. A driving control program for shared electric vehicles was written, and its control functions were implemented using a 51 microcontroller. The Proteus software design interface is shown below, which includes indicator lights, independent button modules, data storage modules, buzzer alarms, IMM-00039 pressure sensors, DS18B20 temperature sensors, C8051F341 microcontrollers, LCD displays, and serial port modules.

[0117] This study employs a convolutional neural network (CNN) architecture to explore the complex task of identifying overloaded shared electric bicycles. A customized model was constructed by fusing multi-scale algorithms, asymmetric convolutional kernels, and a ResNet structure, supplemented by the Adam optimization algorithm and a learning rate decay strategy. The simulation trained on 1000 sets of shared bicycle seat pressure image data covering six core scenarios. Through quantitative evaluation of model performance metrics (accuracy, precision, recall, and F1 score), the following key results and conclusions were obtained:

[0118] 1. Excellent overall performance, significantly surpassing traditional methods: Results: Table 5 shows that the model achieved a high level of recognition accuracy in all scenarios, and the overall model accuracy is expected to reach over 90%.

[0119] Analysis: This result fully demonstrates the superiority of converting sensor data into "pressure images" and applying CNNs for pattern recognition. Compared to the crude judgments of existing technologies that rely on a single gravity sensor or a fixed temperature threshold (which has low accuracy and serious false positives and false negatives), CNN models can learn complex and abstract features from multi-channel (pressure, gradient, dynamic temperature difference) images, thereby achieving a deep understanding of the riding scenario and representing a qualitative leap in performance.

[0120] 2. Significantly improved high similarity scene recognition capability, effectively solving core challenges: Results: For the most challenging “adult + child [overloaded]” (Category 2) and “adult + large object” (Category 4), the model’s accuracy reached 92.3% and 93.55% respectively, and the F1 score also reached 92.25% and 93.14% respectively.

[0121] Analysis: This is one of the most crucial findings of this research. Traditional methods struggle to distinguish between scenes that may be similar in weight but fundamentally different in nature. The CNN model employed in this study, which integrates multi-scale algorithms, asymmetric convolutional kernels, and a ResNet structure, is designed to capture the subtle differences within these "highly similar images." Dynamic temperature difference maps, as a key feature, enable the model to effectively differentiate between living beings with heat sources (children) and inanimate objects without heat sources (backpacks), even if their pressure distributions may be similar. This directly addresses the misclassification and underreporting issues of existing technologies in these critical scenarios, significantly improving the accuracy and reliability of identification.

[0122] 3. Strong environmental adaptability, ensuring all-weather robustness: Results: In the simulation data, whether it is an "empty" or "carrying" scenario, even when the ambient temperature is high in summer, the model can maintain a very high accuracy (e.g., 98.25% for "empty" and 97.4% for "only inanimate objects").

[0123] Analysis: This is thanks to the establishment of a "dynamic temperature baseline" and the application of a "dynamic temperature difference map." The system no longer relies on fixed thresholds, but instead calculates the difference between the actual temperature and the dynamic prediction baseline, effectively eliminating interference from environmental temperature fluctuations. This means that the model can stably and accurately identify human body heat source signals in various environments, including extreme cold and heat, and large diurnal temperature differences, eliminating misjudgments caused by weather changes.

[0124] 4. Strong accurate classification ability, supporting refined management: Results: The model can clearly classify six core scenarios, and all indicators perform well, especially the recognition accuracy of "vacant", "single adult" and "two adults" is over 95%.

[0125] Analysis: This refined scenario classification capability allows the system to go beyond simply determining "whether it is overloaded," and instead understand "the type of overload" and even "whether it is carrying passengers." This provides operators with deeper data insights and offers solid technical support for implementing smarter and more targeted management strategies in the future (e.g., only issuing a warning for goods-carrying vehicles, and forcibly cutting off power for passengers-carrying vehicles).

[0126] Table 5. Accuracy of different scenarios in high-similarity image recognition tasks We randomly selected 100 datasets from the 1000 training datasets (including various categories) for analysis: When a user unlocks a bike, the system immediately collects the initial ambient temperature of the shared bike seat and uploads it to the cloud. The cloud model combines the current time, time zones (which can be correlated with weather data), and historical thermodynamic data of the vehicle to predict a "preset baseline curve" for the temperature change of the shared bike seat over time when the bike is unoccupied. The following are the dynamic baseline curves for temperature prediction in various seasons.

[0127] The following final classification results were obtained by processing and identifying the 100 sets of data selected in the experiment. The seven most representative sets were then selected. The following is a detailed analysis of each set of experimental data:

[0128] 1. Scenario: Empty seat (summer): Pressure map analysis: As shown in Figure 10(a), the pressure of the entire shared bicycle seat is extremely low and uniform, with no obvious concentrated areas, consistent with the empty state. Temperature difference map analysis: The key point is that the ambient baseline temperature is 34.5°C (summer), but the "dynamic temperature difference map" in Figure 10(c) shows a uniform blue color overall, with values ​​close to 0.

[0129] Conclusion: This result further validates the effectiveness of the dynamic environment calibration algorithm. Even under high summer temperatures, the system can accurately determine that there are no "abnormal heat sources" on the shared bicycle seats, thus avoiding false alarms that may be caused by traditional fixed threshold schemes.

[0130] 2. Scenario: Single adult, normal weight (summer): Pressure map analysis: As shown in Figure 11(a), a typical single-person riding pattern is presented: an elliptical area with high central pressure that spreads outwards; Gradient map analysis: As shown in Figure 11(b), the outline and edge gradient of the pressure area of ​​a single human body show a lower gradient in the central area (gradual pressure) and a higher gradient at the edges (pressure variation); Temperature difference map analysis: As shown in Figure 11(c), a significant red heat source signal appears in the center of the pressure area, clearly indicating the presence of a life form above the environmental baseline in this area.

[0131] Conclusion: Through the collaborative work of multiple features, the model can easily identify the pattern of "single pressure area + single heat source" and accurately classify it as "single adult".

[0132] 3. Scenario: Adults + Children (Autumn, Overload): Pressure Map Analysis: As shown in Figure 12(a), a large, slightly pear-shaped pressure area is displayed, with high overall pressure intensity, indicating a load of more than one person; Gradient Map Analysis: As shown in Figure 12(b), the gradient map has a more complex structure in the center, reflecting two overlapping stress points; Temperature Difference Map Analysis: This is the core of identifying overload. As shown in Figure 12(c), two significant red heat source signals are clearly displayed. They partially overlap, and their shape and intensity indicate the presence of two independent life forms.

[0133] Conclusion: By identifying the key pattern of "complex pressure distribution + two independent / overlapping heat sources", the model accurately classifies "adult + child" overload, overcoming the challenge of distinguishing based solely on total weight.

[0134] 4. Scenario: Two adults (summer, overloaded): Pressure map analysis: As shown in Figure 13(a), a very wide, high-intensity pressure area with possibly two peaks is displayed, almost occupying the entire shared bicycle seat, with generally high pressure values; Gradient map analysis: As shown in Figure 13(b), two obvious, moderately spaced black "indentations" (corresponding to the pressure center) and surrounding red high gradient areas appear in the center of the gradient map, reflecting the existence of two stress points; Temperature difference map analysis: As shown in Figure 13(c), two large-scale, high-intensity red heat source signals are also presented, and their distribution matches the pressure area, further confirming the existence of two living beings.

[0135] Conclusion: Wide-ranging, high-intensity, bi-peak pressure and temperature characteristics are clear signals of “two adults” overload, and the model can identify them efficiently.

[0136] 5. Key Scenario: Adult + Object (Summer): Pressure Map Analysis: As shown in Figure 14(a), two distinct pressure zones are observed, one elliptical and located further back (human body), and the other more irregularly shaped and located further forward (object). Gradient Map Analysis: Figure 14(b) better reveals this separation and shape difference; for example, the edge gradient of the object area may be sharper. Temperature Difference Map Analysis: This is crucial for distinguishing this scenario from "Adult + Child." Despite an ambient baseline temperature of 29.0°C (summer), the "Dynamic Temperature Difference Map" in Figure 14(c) shows only one significant red heat source signal, located in the human body area at the rear of the shared bicycle seat. The object area at the front, regardless of whether it is heated by the sun, has a temperature difference close to zero from the ambient baseline, generating no additional heat source signal.

[0137] Conclusion: The results demonstrate how the system can accurately distinguish between two highly similar scenarios, "adult + object" and "adult + child," by using a combination of "two pressure zones + only one heat source signal," thus avoiding misjudgment.

[0138] 6. Scenario: Inanimate Objects Only (Summer): Pressure Map Analysis: As shown in Figure 15(a), one or more obvious pressure areas are displayed (here, two relatively independent small areas simulating two small objects), but their shapes do not conform to human biomechanical characteristics; Gradient Map Analysis: As shown in Figure 15(b), the gradient map clearly shows the outlines of these two objects; Temperature Difference Map Analysis: This is the decisive evidence. Although the ambient baseline temperature is 28.8°C (summer), the "Dynamic Temperature Difference Map" in Figure 15(c) presents a uniform blue color overall, with no red heat source signals detected.

[0139] Conclusion: This further demonstrates the powerful ability of dynamic temperature difference maps to eliminate interference from non-living objects. The system can determine with 100% certainty that this is not a manned scenario, thus effectively avoiding false alarms caused by placing heavy objects and improving the user experience.

[0140] Overall experimental conclusions: Simulation data further reinforces previous findings, demonstrating that this technical solution successfully achieves high-precision and robust recognition of shared electric bicycle riding scenarios by converting multi-channel sensor data into "pressure images" and combining them with dynamically adaptive temperature difference maps. In particular, its superior performance in distinguishing between highly similar overloaded and cargo-carrying scenarios such as "adult + child" and "adult + object," as well as its accurate judgment of "empty" and "carrying only" scenarios under high-temperature environments, fully validates the significant advantages and reliability of this solution in practical applications.

[0141] This invention proposes an intelligent overload recognition system for shared bicycles, comprising: a data acquisition module for collecting temperature and pressure data of the shared bicycle seat using a sensor array installed within the seat; an image conversion module for reshaping multiple discrete temperature and pressure data into two independent matrices to obtain an original pressure map and an original temperature map; a three-channel module for constructing a three-channel input tensor using the original pressure map and the original temperature map; the three-channel input tensor includes: a normalized pressure map, a pressure gradient map obtained based on the normalized pressure map, and a dynamic temperature difference map obtained based on the original temperature map and a dynamic environmental reference temperature, the dynamic environmental reference temperature being predicted based on historical data; and a recognition module for performing pattern recognition on the multiple input tensors to obtain classification results belonging to different riding scenarios, and controlling the shared bicycle to perform corresponding operations based on the classification results, wherein when the classification result is an overload scenario, an alarm is triggered or the vehicle operation is restricted.

[0142] The present invention also provides a computer device, including a memory and a processor. The memory stores a program, and when the program is executed by the processor, the processor performs the steps of a method for intelligent identification of overloaded shared bicycles.

[0143] According to the disclosed embodiments, the computer device can communicate with one or more external devices (e.g., keyboard, pointing device, Bluetooth communication, etc.) or with any device that enables the computing device to communicate with one or more other computing devices (e.g., router, demodulator, etc.).

[0144] The present invention also provides a storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of a method for intelligent identification of overloaded shared bicycles.

[0145] According to the disclosed embodiments, the storage medium can be a non-volatile computer-readable storage medium, such as, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, the storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0146] The above description, in conjunction with specific preferred embodiments, provides a more detailed explanation of the present invention. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection of the present invention.

Claims

1. A method for intelligent identification of overloaded shared bicycle riders, characterized in that, The steps include: collecting temperature and pressure data from the shared bicycle seat using a sensor array within the seat; reshaping multiple discrete temperature and pressure data into two independent matrices to obtain the original pressure map and the original temperature map; A three-channel input tensor is constructed using the original pressure map and the original temperature map; The three-channel input tensors include: a normalized pressure map, a pressure gradient map obtained based on the normalized pressure map, and a dynamic temperature difference map obtained based on the original temperature map and the dynamic environmental reference temperature, wherein the dynamic environmental reference temperature is predicted based on historical data. Pattern recognition is performed on multiple input tensors to obtain classification results belonging to different riding scenarios, and the shared bicycles are controlled to perform corresponding operations according to the classification results. When the classification result is an overload scenario, an alarm is triggered or the vehicle operation is restricted.

2. The intelligent identification method for overloading of shared bicycles as described in claim 1, characterized in that, The dynamic temperature difference map, obtained based on the original temperature map and the dynamic environmental reference temperature, specifically involves: when a user unlocks the vehicle, the initial environmental temperature of the shared bicycle seat is collected and uploaded to the cloud; the cloud model, combining time, geographical location information, and historical thermodynamic data of the vehicle, predicts a dynamic environmental reference temperature curve showing the temperature change of the shared bicycle seat over time when the vehicle is unoccupied; during real-time detection, the difference between the temperature value at each point in the original temperature map and the corresponding dynamic environmental reference temperature curve is obtained to generate the dynamic temperature difference map.

3. The intelligent identification method for overloading of shared bicycles as described in claim 1, characterized in that, The process of performing pattern recognition on multiple input tensors to obtain classification results belonging to different cycling scenarios specifically involves: inputting each input tensor into the convolutional layer of a convolutional neural network model, and passing it through a transistor of size [size missing]. convolution kernel The image is slid across the image to perform a dot product operation, resulting in an output feature map. The output feature map is then max-pooled and flattened into a one-dimensional vector. This one-dimensional vector is then mapped to a space with dimensions below a threshold through a fully connected layer. The output of the fully connected layer is converted into a probability distribution vector, and the classification results for different cycling scenarios are obtained through this probability distribution vector.

4. The intelligent identification method for overloading of shared bicycles as described in claim 3, characterized in that, The convolutional neural network model is trained using sample data containing various cycling scenarios, which include at least: an empty state, a single adult cycling state, an overloaded state with an adult carrying a child, an overloaded state with two adults, a state with an adult carrying an object, and a state with only inanimate objects placed on the ground.

5. The intelligent identification method for overloading of shared bicycles as described in claim 4, characterized in that, The convolutional neural network model is constructed using a fusion of multi-scale algorithms, asymmetric convolutional kernels, and residual network structures, and trained using the Adam optimization algorithm and a learning rate decay strategy.

6. The intelligent identification method for overloading of shared bicycles as described in claim 1, characterized in that, When collecting the temperature and pressure data of the shared bicycle seat, the specific process is as follows: the digital signal output by the sensor is converted into degrees Celsius, and the specific expression is: ;in, In the sensor array The actual temperature value at the location, This is the raw digital reading read from the position sensor; the collected resistance change is converted into a pressure value, specifically expressed as: ;in, Is The pressure value at the location, It is the measured change in resistance. It is the sensor sensitivity coefficient that needs to be calibrated through experiments.

7. The intelligent identification method for overloading of shared bicycles as described in claim 1, characterized in that, The pressure sensor is an IMM-00039 series distributed multi-point pressure sensing resistive sensor, and the temperature sensor is a DS18B20 temperature sensor.

8. A smart recognition system for overloading of shared bicycles, characterized in that, include: The data acquisition module is used to collect the temperature and pressure of the shared bicycle seat through a sensor array inside the shared bicycle seat. The image conversion module is used to reshape multiple discrete temperatures and pressures into two independent matrices to obtain the original pressure map and the original temperature map. A three-channel module is used to construct a three-channel input tensor from the original pressure map and the original temperature map; The three-channel input tensors include: a normalized pressure map, a pressure gradient map obtained based on the normalized pressure map, and a dynamic temperature difference map obtained based on the original temperature map and the dynamic environmental reference temperature, wherein the dynamic environmental reference temperature is predicted based on historical data; the recognition module is used to perform pattern recognition on multiple input tensors to obtain classification results belonging to different riding scenarios, and control the shared bicycle to perform corresponding operations according to the classification results, wherein when the classification result is an overload scenario, an alarm is triggered or the vehicle operation is restricted.

9. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores a program, and when the program is executed by the processor, the processor performs the steps of the intelligent identification method for overloading of shared bicycles as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent identification method for overloaded shared bicycle riding as described in any one of claims 1 to 7.

Images