Sanitation vehicle suction nozzle self-adaptive adjusting system and method based on AI visual identification

Abstract

The invention discloses a sanitation vehicle suction nozzle adaptive adjustment system and method based on AI visual identification, and the method comprises the steps: a), obtaining RGB image data of road garbage through a visual sensor, detecting the garbage accumulation height through a laser radar, and collecting the real-time speed of a vehicle in combination with an inertial measurement unit (IMU); b) processing the RGB image data by using a deep learning model, identifying a garbage type, outputting a category label and confidence, calculating a garbage volume based on a depth map, and generating a two-dimensional feature vector containing the garbage type and the volume; c) matching a reference suction parameter in a preset database according to the garbage type, and calculating a compensation suction value in combination with the real-time speed of the vehicle, and d) adjusting the rotating speed of a turbine fan through a fan controller, and controlling the pressure of a suction nozzle within a target suction value range.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent environmental sanitation equipment, and particularly relates to a sanitation vehicle suction nozzle adaptive adjustment system and method based on AI visual recognition. Background Art

[0002] The following problems exist in the suction adjustment of traditional sanitation vehicles:

[0003] Fixed suction mode: It cannot dynamically adjust according to the type of garbage (such as light leaves and heavy metal cans), resulting in energy consumption waste or incomplete recycling.

[0004] Sensor limitations: Existing solutions rely on infrared sensors to detect the volume of garbage and cannot identify material differences, with a high misjudgment rate.

[0005] Response delay: The visual recognition technology and the suction control system are not deeply coupled, and the response time exceeds 500 ms, making it difficult to meet the real-time operation requirements. Summary of the Invention

[0006] The purpose of the present invention is to provide a sanitation vehicle suction nozzle adaptive adjustment system and method based on AI visual recognition, in order to solve the technical problems existing in the background art.

[0007] In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0008] A sanitation vehicle suction nozzle adaptive adjustment method based on AI visual recognition, comprising the following steps:

[0009] a) Obtain RGB image data of road garbage through a visual sensor, detect the garbage accumulation height through a lidar, and collect the vehicle's real-time speed in combination with an inertial measurement unit (IMU);

[0010] b) Process the RGB image data using a deep learning model to identify the type of garbage and output the class label and confidence level. At the same time, calculate the garbage volume based on the depth map and generate a two-dimensional feature vector containing the garbage type and volume;

[0011] c) Match the benchmark suction parameters in the preset database according to the type of garbage, and calculate the compensation suction value in combination with the vehicle's real-time speed, where the compensation suction value satisfies the formula:

[0012] Q = Q_base × (1 + kv),

[0013] where Q_base is the benchmark suction, k is the adjustment coefficient and satisfies 0 < k ≤ 0.3, and v is the vehicle speed;

[0014] d) Adjust the turbine fan speed through the fan controller to control the suction nozzle pressure within the target suction value range, with an error not exceeding ±2 kPa.

[0015] In some embodiments, the visual sensor in step a) is a binocular camera with a frame rate of not less than 30 fps and supports image acquisition under complex lighting conditions.

[0016] In some embodiments, the deep learning model in step b) is trained through data enhancement, and the training data includes synthetic data of rainy day reflections, low light at night, and mixed scenes of multiple types of garbage.

[0017] In some embodiments, the preset database in step c) stores a mapping relationship between garbage types and reference suction parameters, including plastic bottles → 35 kPa, metal cans → 50 kPa, and leaves → 20 kPa.

[0018] In some embodiments, it also includes: detecting the garbage accumulation height through laser radar, and when the height exceeds a preset threshold, triggering the automatic adjustment mechanism of the suction nozzle height to adapt to the garbage accumulation shape.

[0019] In some embodiments, step d) further includes: predicting the garbage distribution within the 5-meter road section ahead based on the LSTM neural network, and pre-adjusting the suction parameters according to the prediction results; automatically switching to low power consumption mode in garbage-free sections to maintain a basic suction of 10kPa.

[0020] In some embodiments, the method further includes dynamically adjusting the opening of the suction nozzle valve according to the volume of the garbage, and increasing the valve opening when the volume of the garbage exceeds a set threshold to improve the garbage suction efficiency.

[0021] This embodiment also provides an adaptive adjustment system for a sanitation vehicle nozzle based on AI visual recognition, including:

[0022] a) Visual perception module, including binocular cameras and edge computing units, is used to collect and process road garbage image data in real time, supporting classification and identification of more than 15 types of garbage with an accuracy rate of ≥ 93%;

[0023] b) Suction control actuator, including a variable frequency fan and an air pressure sensor, forming a closed-loop control system for accurately adjusting the suction nozzle pressure;

[0024] c) Onboard central controller, deploying a fusion decision-making algorithm with a system response time of ≤200ms, used to integrate multi-source data and generate suction adjustment instructions;

[0025] d) Human-machine interface (HMI), used to display garbage identification results, real-time suction value and energy consumption curve.

[0026] The camera of the visual perception module has an IP67 protection level, and the edge computing unit is integrated into the vehicle controller.

[0027] The on-board central controller dynamically optimizes the fan power according to the vehicle's speed and the garbage distribution prediction results, thereby reducing energy consumption by more than 15%.

[0028] The beneficial effects of the adaptive adjustment system and method for sanitation vehicle nozzles based on AI visual recognition disclosed in this application may include but are not limited to:

[0029] Improved waste recycling rate: Through precise classification and dynamic suction adjustment, the recycling rate of light waste has increased to 98%, and the recycling rate of heavy waste has increased to 95%;

[0030] Energy consumption optimization: comprehensive energy saving of more than 15%, and additional energy saving of 30% in low power mode;

[0031] Real-time guarantee: The system response time is ≤200ms, which is suitable for continuous operation at a vehicle speed of 20km / h. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic diagram of the system of this application. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0034] On the contrary, this application covers any alternatives, modifications, equivalents, and solutions made within the spirit and scope of this application as defined by the claims. Furthermore, to facilitate a better understanding of this application, certain specific details are described in detail below in the detailed description of this application. Those skilled in the art will be able to fully understand this application without these details.

[0035] A method for adaptively adjusting a sanitation vehicle nozzle based on AI visual recognition includes the following steps:

[0036] c) Obtain RGB image data of road garbage through visual sensors, detect garbage accumulation height through lidar, and collect real-time vehicle speed in combination with an inertial measurement unit (IMU);

[0037] d) processing the RGB image data using a deep learning model to identify the type of garbage and output a category label and confidence score, while calculating the volume of the garbage based on the depth map to generate a two-dimensional feature vector containing the garbage type and volume;

[0038] c) Matching the baseline suction parameters in the preset database according to the garbage type and calculating the compensation suction value in combination with the real-time speed of the vehicle, wherein the compensation suction value satisfies the formula:

[0039] Q = Q_base×(1 + kv),

[0040] where Q_base is the reference suction, k is the adjustment coefficient and satisfies 0 < k ≤ 0.3, and v is the vehicle speed;

[0041] d) Adjust the rotational speed of the turbine fan through the fan controller to control the nozzle pressure within the target suction value range, with an error not exceeding ±2 kPa.

[0042] In some embodiments, in step a), the vision sensor is a binocular camera with a frame rate not lower than 30 fps and supports image acquisition under complex lighting conditions.

[0043] In some embodiments, in step b), the deep learning model is trained through data augmentation, and the training data includes synthetic data of rainy-day reflection, nighttime low-light, and multi-type garbage mixed scenarios.

[0044] In some embodiments, in step c), the preset database stores the mapping relationship between garbage types and reference suction parameters, including plastic bottle → 35 kPa, metal can → 50 kPa, leaf → 20 kPa.

[0045] In some embodiments, it further includes: detecting the garbage accumulation height through a lidar, and when the height exceeds a preset threshold, triggering an automatic nozzle height adjustment mechanism to adapt to the garbage accumulation form.

[0046] In some embodiments, step d) further includes: predicting the garbage distribution within a 5-meter section ahead based on the LSTM neural network and pre-adjusting the suction parameters according to the prediction results; automatically switching to a low-power consumption mode in a garbage-free section and maintaining a 10 kPa basic suction.

[0047] In some embodiments, it further includes dynamically adjusting the opening degree of the nozzle valve according to the garbage volume, and increasing the valve opening degree to improve the garbage suction efficiency when the garbage volume exceeds the set threshold.

[0048] As Figure 1 shown, this embodiment also provides an AI vision recognition-based adaptive adjustment system for the suction nozzle of a sanitation vehicle, including:

[0049] a) A vision perception module, including a binocular camera and an edge computing unit, for real-time collecting and processing road surface garbage image data, supporting the recognition of more than 15 types of garbage, with an accuracy rate ≥ 93%;

[0050] b) A suction control actuator, including a variable-frequency fan and a pressure sensor, constituting a closed-loop control system for precisely adjusting the nozzle pressure;

[0051] c) Onboard central controller, deploying a fusion decision-making algorithm with a system response time of ≤200ms, used to integrate multi-source data and generate suction adjustment instructions;

[0052] d) Human-machine interface (HMI), used to display garbage identification results, real-time suction value and energy consumption curve.

[0053] The camera of the visual perception module has an IP67 protection level, and the edge computing unit is integrated into the vehicle controller.

[0054] The on-board central controller dynamically optimizes the fan power according to the vehicle's speed and the garbage distribution prediction results, thereby reducing energy consumption by more than 15%.

[0055] Take the municipal road sweeper as an example

[0056] Hardware deployment:

[0057] Visual perception module:

[0058] Two 2-megapixel binocular cameras (30fps, IP67 protection level) are installed on the front of the vehicle, covering a 120° horizontal viewing angle;

[0059] The edge computing unit is integrated into the vehicle controller and supports TensorRT accelerated reasoning.

[0060] Suction control actuator:

[0061] A high-precision air pressure sensor is integrated inside the nozzle (range 0-100kPa, error ±0.5%);

[0062] The variable frequency fan adopts a permanent magnet synchronous motor with a speed range of 500-3000rpm and a response time of ≤0.2s.

[0063] On-board central controller:

[0064] Based on the ARM Cortex-A72 processor, running a real-time operating system (RTOS);

[0065] Deploy a multi-threaded fusion decision algorithm to synchronously process image, radar and IMU data.

[0066] Algorithm implementation:

[0067] 1. Data enhancement and model training:

[0068] The original dataset contained 100,000 garbage images (covering 15 common types of garbage), which was expanded to 300,000 by synthesizing scenes such as rainy day reflections and nighttime low-light scenes.

[0069] Using a transfer learning strategy, the classification model was fine-tuned based on the pre-trained ResNet-50, and the accuracy of the test set reached 93.5%.

[0070] 2. Real-time control logic:

[0071] The vehicle speed is updated every 0.1s, and the suction compensation coefficient k is set to 0.2 (the optimal value verified by experiments);

[0072] When the laser radar detects that the garbage accumulation height exceeds 30cm, the suction nozzle height will automatically rise by 5cm.

[0073] Typical scenarios:

[0074] A plastic bottle (volume 0.5L, confidence level 0.92) is detected 3m ahead:

[0075] Call the basic suction parameter 35kPa;

[0076] Based on the current vehicle speed of 8 km / h (≈2.22 m / s), the compensation suction force Q = 35 × (1 + 2.22 / 10) = 42.77 kPa;

[0077] The fan controller increases the fan speed to 2200 rpm within 0.15 s.

[0078] Scenario 2: Metal Can Recycling

[0079] A metal can (volume 1.2L, confidence level 0.95) was detected 3m ahead.

[0080] The reference suction is 50 kPa, the current vehicle speed is 10 km / h (≈2.78 m / s), and the compensation suction is calculated as Q = 50 × (1 + 0.2 × 2.78) = 57.8 kPa;

[0081] The fan controller increased the speed to 2500 rpm within 0.15 s, and the nozzle pressure was stabilized at 57.8 ± 2 kPa.

[0082] Scenario 3: Low power mode switching

[0083] LSTM predicts that there is no garbage distribution in the 5m ahead;

[0084] The central controller sends a command to reduce the suction force to 10kPa and adjust the fan speed to 800rpm, reducing energy consumption by 60%.

[0085] The AI-based visual recognition-based adaptive adjustment system and method for sanitation vehicle nozzles provided by this invention achieves the following significant advantages through multimodal perception fusion, dynamic suction control, and intelligent energy consumption optimization:

[0086] Garbage collection efficiency has been significantly improved

[0087] An improved deep learning model (with an accuracy rate of ≥93%) accurately identifies waste type and volume, combined with type-matched baseline suction parameters (e.g., 50kPa for cans, 20kPa for leaves), effectively avoiding the "undersuction" or "overconsumption" issues associated with traditional fixed suction modes. Experiments have shown that the recovery rate for lightweight waste (such as leaves) has increased to 98%, and the recovery rate for heavy waste (such as cans) has increased to 95%.

[0088] Dynamic optimization of energy consumption, outstanding energy-saving effect

[0089] By introducing a speed compensation factor (Q = Q_base × (1 + kv)) and a low-power mode (maintaining a base suction force of 10 kPa), combined with an LSTM neural network to predict the distribution of trash ahead, the system can adjust suction force on demand. Actual measured data shows a reduction of over 15% in overall energy consumption, with energy savings reaching 30% on trash-free roads.

[0090] Enhanced real-time performance and adaptability

[0091] Through the edge computing unit (deployed in the vehicle controller) and closed-loop control technology, the system response time is shortened to ≤200ms, meeting the real-time operation requirements at a vehicle speed of 20km / h.

[0092] The improved visual algorithm supports garbage recognition in complex environments (such as at night and in rainy and reflective environments). The test set covers scenes such as fog and shade, and its robustness is significantly improved.

[0093] High level of automation and intelligence

[0094] The laser radar triggers the automatic adjustment of the suction nozzle height (for example, it will be raised by 5 cm when the garbage accumulation height exceeds 30 cm). The opening of the suction nozzle valve is dynamically adjusted according to the volume of garbage, reducing manual intervention and improving operation safety.

[0095] The HMI interface displays waste identification results, suction parameters and energy consumption curves in real time, making it easier for operators to monitor and optimize operating strategies.

[0096] Broad prospects for industrial applications

[0097] This system can be directly adapted to existing municipal sweepers, airport cleaning vehicles and other equipment, assisting smart city sanitation management and promoting the realization of green and low-carbon goals.

[0098] 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 in the scope of protection of the present invention.

Claims

1. A self-adaptive adjustment method for a sanitation vehicle nozzle based on AI visual recognition, characterized in that: It includes the following steps: a) Obtain the RGB image data of road surface garbage through a vision sensor, detect the garbage accumulation height through a lidar, and collect the vehicle's real-time speed in combination with an Inertial Measurement Unit (IMU); b) Process the RGB image data using a deep learning model to identify the garbage type and output the class label and confidence level. At the same time, calculate the garbage volume based on the depth map and generate a two-dimensional feature vector containing the garbage type and volume; c) Match the benchmark suction parameter in the preset database according to the garbage type, and calculate the compensation suction value in combination with the vehicle's real-time speed, where the compensation suction value satisfies the formula: Q = Q_base×(1 + kv), where Q_base is the benchmark suction, k is the adjustment coefficient and satisfies 0 < k ≤ 0.3, and v is the vehicle speed; d) Adjust the rotational speed of the turbo fan through the fan controller to control the nozzle pressure within the target suction value range.

2. The method according to claim 1, wherein In step a), the vision sensor is a binocular camera with a frame rate not lower than 30fps and supports image acquisition under complex lighting conditions.

3. The method according to claim 1, wherein In step b), the deep learning model is trained through data augmentation, and the training data includes synthetic data of rainy-day reflection, low-light at night, and mixed scenarios of multiple types of garbage.

4. The method according to claim 1, wherein In step c), the preset database stores the mapping relationship between the garbage type and the benchmark suction parameter, including plastic bottle → 35kPa, metal can → 50kPa, leaf → 20kPa.

5. The method according to claim 1, wherein It also includes: Detect the garbage accumulation height through a lidar. When the height exceeds the preset threshold, trigger the automatic adjustment mechanism of the nozzle height to adapt to the garbage accumulation form.

6. The method according to claim 1, wherein Step d) further includes: Predict the garbage distribution within a preset length of the road ahead based on the LSTM neural network, and pre-adjust the suction parameter according to the prediction result; automatically switch to the low-power mode in the section without garbage to maintain the basic suction.

7. The method according to claim 1, wherein It also includes: Dynamically adjust the opening degree of the nozzle valve according to the garbage volume. When the garbage volume exceeds the set threshold, increase the valve opening degree to improve the garbage suction efficiency.

8. An adaptive adjustment system for the suction nozzle of a sanitation vehicle based on AI visual recognition, characterized in that: It includes: a) A visual perception module, including a binocular camera and an edge computing unit, for real-time collecting and processing the road surface garbage image data; b) A suction control actuator, including a variable-frequency fan and a pressure sensor, constituting a closed-loop control system for precisely adjusting the nozzle pressure; c) A vehicle-mounted central controller for integrating multi-source data and generating a suction adjustment instruction; d) A Human-Machine Interface (HMI) for displaying the garbage recognition result, real-time suction value, and energy consumption curve.

9. The system according to claim 8, wherein The camera of the visual perception module has an IP67 protection level, and the edge computing unit is integrated into the vehicle's whole vehicle controller.

10. The system according to claim 8, wherein The vehicle-mounted central controller dynamically optimizes the fan power according to the vehicle's traveling speed and the garbage distribution prediction result.

Images