A portable handheld terminal for detecting the anticorrosion surface quality of a hydraulic metal structure

Abstract

This invention discloses a portable handheld terminal for inspecting the surface quality of corrosion-resistant hydraulic metal structures, belonging to the field of hydraulic metal structure quality inspection technology. The terminal includes an integrated handheld protective carrier, a core main control module, an image acquisition and recognition system, a field lighting component, a human-computer interaction touch unit, an end-side intelligent processing unit, and a power supply unit. These units are integrated and electrically coordinated. The core control unit is equipped with an embedded Linux system, and the end-side pre-stores a visual intelligent detection algorithm model optimized by a lightweight module, coupled with NPU hardware acceleration, enabling real-time offline inference analysis. This allows for integrated intelligent detection of rust grading, coating defects, coating adhesion, rust removal cleanliness, and roughness. The terminal features a single-handed grip, a waterproof and dustproof structure, supports non-contact inspection and visual human-computer interaction, and can adapt to complex operating scenarios in the field with no network and high humidity and dust. It effectively solves the problems of existing inspection equipment being single-function, inefficient, and highly subjective, and features high integration, portability, stability, intelligence, and efficiency.

Description

Technical Field

[0001] This invention relates to the field of quality inspection technology for hydraulic metal structures, specifically to a portable handheld terminal for inspecting the anti-corrosion surface quality of hydraulic metal structures. It is suitable for inspecting the anti-corrosion surface quality of hydraulic metal structures such as gates, trash racks, and pressure steel pipes in water conservancy projects. It can perform core inspection tasks such as coating defect identification, rust level determination, coating adhesion determination, surface cleanliness determination, and visual evaluation of surface roughness. It is especially suitable for complex working conditions such as damp, dusty, and unconnected outdoor environments. Background Technology

[0002] As core load-bearing components of hydraulic engineering projects, hydraulic metal structures operate long-term in harsh environments characterized by underwater conditions, high humidity, dust, and large temperature fluctuations. The quality of their surface anti-corrosion treatment directly determines the structure's service life and operational safety. Anti-corrosion quality testing of hydraulic metal structures must cover all key stages of the process: after rust removal from the substrate, surface cleanliness and roughness must be tested to ensure compliance with coating application requirements; after coating application, coating defects and adhesion must be tested to verify construction quality; during routine operation and maintenance, the substrate's corrosion status must be monitored regularly to promptly identify potential anti-corrosion failures. Failure to meet any of these testing requirements can lead to coating peeling, accelerated structural corrosion, and even serious engineering accidents such as structural fractures and leaks.

[0003] Currently, existing methods for inspecting the surface quality of corrosion protection on hydraulic metal structures have many prominent problems and are difficult to meet actual on-site needs:

[0004] Fragmented testing process: Existing testing equipment has limited functions. Core indicators such as coating defects, degree of corrosion, cleanliness, and roughness need to be tested by multiple independent devices in multiple tests. A large number of tools need to be carried on site, making the operation cumbersome, time-consuming, and labor-intensive. Especially when testing in special locations such as high altitudes or near water, the difficulty of moving and switching equipment is extremely high, resulting in very low testing efficiency.

[0005] The test results are highly subjective: Traditional testing relies on manual experience and lacks unified quantitative standards. Different operators have different judgment results, which can easily lead to misjudgment and missed detection, resulting in rework of coating construction or failure of anti-corrosion potential in a timely manner. At the same time, manual testing is difficult to retain original images and video data, making subsequent traceability and archiving difficult.

[0006] Poor equipment adaptability: Most testing equipment is not designed for hydraulic engineering field environments and does not have waterproof and dustproof performance, making it prone to failure in humid and dusty environments; some equipment is large and heavy, lacking portability and unable to adapt to scenarios such as confined spaces and high-altitude operations; moreover, some equipment relies on networks and cloud computing, and cannot be used in environments without network access, making it impossible to obtain test results in real time on site.

[0007] Inconvenient operation and battery life: Existing equipment lacks user-friendly interactive design, has complex parameter configuration, and requires professional training to operate; some equipment has weak battery life, which cannot meet the needs of long-term on-site operation, and the data storage and export process is cumbersome, affecting the continuity of testing work.

[0008] To address the aforementioned technical challenges, there is an urgent need to develop an integrated, portable, and intelligent handheld testing terminal that enables non-contact, real-time, and accurate testing of multiple indicators on the anti-corrosion surface of hydraulic metal structures. This would simplify the testing process, improve testing efficiency and standardization, and be adaptable to complex field operating environments. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the prior art and provide a portable handheld terminal for detecting the surface quality of anti-corrosion metal structures in hydraulic engineering. This terminal solves the technical problems of existing testing equipment having limited functions, cumbersome testing procedures, highly subjective results, poor adaptability, and inconvenient operation. It enables integrated, non-contact, end-side offline real-time detection of the surface quality of anti-corrosion metal structures in hydraulic engineering, while taking into account portability, stability, and detection accuracy.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] A portable handheld terminal for inspecting the surface quality of corrosion protection on hydraulic metal structures is disclosed. The handheld terminal is a non-contact visual inspection structure, comprising an integrated handheld protective carrier, a core control module, an image acquisition and recognition component, an on-site lighting component, a human-machine interaction touch component, and a power supply module. The core control module, image acquisition and recognition component, on-site lighting component, human-machine interaction touch component, and power supply module are all integrated and installed inside the integrated handheld protective carrier. Each functional component is electrically connected to the core control module and is uniformly scheduled and controlled by the core control module. The core control module locally pre-stores a lightweight YOLOv8 visual inspection algorithm model, integrates end-side intelligent processing functions, and communicates with the image acquisition and recognition component. It is used to receive surface image acquisition data, perform intelligent analysis of corrosion protection quality based on the surface characteristics of the hydraulic metal structure, and synchronize the analysis results to the human-machine interaction touch component.

[0012] Furthermore, the integrated handheld protective carrier features a single-handed ergonomic shell structure with an overall protection level of IP56, providing dust and water resistance. This effectively protects against the humid and dusty environment of hydraulic engineering sites. The shell has an internal active heat dissipation duct that circulates air to cool the core electronic components, preventing overheating failures caused by prolonged high-load operation. The external data transmission and charging interfaces are equipped with waterproof sealing rings and anti-loosening buckles, further enhancing the sealing and protection effect and preventing moisture and dust from entering the device.

[0013] Furthermore, the core main control module adopts an integrated embedded main control development board, which is equipped with a multi-core processor, a graphics processing unit, and a neural network computing unit. It runs on an embedded operating system based on the Linux kernel, which is stable and has low power consumption. The core main control module has the ability to perform hardware power-on self-test, start and stop scheduling of functional components, and full-process control of detection. After powering on, it can automatically complete the status detection and initialization of each component without manual intervention. It also integrates a wireless communication unit and a wired communication unit to realize bidirectional data transmission between the terminal and external devices, taking into account the data transmission needs of different scenarios.

[0014] Furthermore, the wireless communication unit of the core main control module includes a Wi-Fi communication unit and a Bluetooth communication unit, which can realize convenient wireless data transmission and device interconnection; the wired communication unit adopts a gigabit Ethernet communication unit, combined with a USB 3.0 interface, which can meet the high-speed export requirements of a large amount of test data, and take into account both convenient wireless transmission and high-speed wired data export scenarios on site.

[0015] Furthermore, the image acquisition and recognition component is fixedly mounted on the front detection side of the integrated handheld protective carrier. It adopts an industrial-grade vision acquisition module, which integrates automatic focus adjustment, image parameter (gain, exposure time) adjustment, and white balance correction functions. It can accurately capture the micron-level morphological features and true color information of the anti-corrosion surface of hydraulic metal structures. The component supports the acquisition of high-definition still images with a resolution of 3840×2160 and real-time dynamic video at 60fps. The acquired data is transmitted to the core main control module in real time through a high-speed USB 3.0 interface, providing a high-quality data source for subsequent analysis.

[0016] Furthermore, the on-site supplementary lighting component is arranged on the same side as the image acquisition and recognition component, and the supplementary lighting optical path is coaxially aligned with the image acquisition optical path to ensure accurate coverage of the detection area. The on-site supplementary lighting component adopts a constant brightness high-intensity LED light source module with a rated power supply voltage of 24V. It can automatically adjust the output brightness according to the ambient light intensity, effectively making up for insufficient ambient light and avoiding imaging interference caused by reflections or shadows on metal surfaces, thus ensuring uniform brightness and clear details in the image acquisition screen.

[0017] Furthermore, the human-machine interaction touch control component is located on the front operating area of ​​the integrated handheld protective carrier, and adopts a 5-inch industrial-grade anti-interference touch display screen with fast response and clear display in both strong light and dark environments. The display screen is equipped with a visual interactive interface, which is divided into a parameter configuration area, a real-time screen area, a detection result area, and a function control area. It supports the configuration of detection parameters, switching of detection modes, real-time acquisition screen display, and visualization of detection results. The operation logic is simple and intuitive.

[0018] Furthermore, the core control module, relying on its built-in neural network computing unit, optimizes computing power and model adaptation for corrosion detection scenarios of hydraulic metal structures, enabling real-time offline inference computation on the edge without relying on cloud computing or network support. The development board pre-stores lightweight visual detection algorithm models locally, and through lightweight processing such as model pruning and quantization, it adapts to the embedded hardware computing power, enabling rapid intelligent analysis and grade determination of corrosion-resistant surface quality. The determination data is synchronized to the human-computer interaction touch component in real time, providing real-time display from image acquisition to result output.

[0019] Furthermore, the core main control unit locally pre-stores lightweight optimized visual intelligent detection algorithm models. The algorithm models include a rust level segmentation model, a coating defect target detection model, a coating adhesion classification model, a cleanliness classification model, and a roughness classification model. All models use a lightweight feature extraction module to replace the native C2f feature extraction module in the trunk and neck structure of the YOLOv8 model. The module as a whole is composed of a front 1×1 pointwise convolutional channel compression layer, a dual-branch parallel feature processing structure, and a rear 1×1 pointwise convolutional feature fusion output layer connected in series.

[0020] Furthermore, the core technology of the pre-convolutional channel compression layer (1×1 pointwise convolutional channel compression layer) is to perform linear compression processing on the input feature map along the channel dimension without changing the spatial resolution of the feature map. This process removes redundant and invalid information within the channel dimension from the source, effectively reducing the computational load and parameter consumption of the subsequent dual-branch structure. Its convolution operation formula is as follows: In the formula, Input the original feature map into the module. It is a 1×1 pointwise convolution kernel weight matrix. For the corresponding convolutional layer bias term, Using a linear activation function avoids the loss of feature information caused by nonlinear transformations and preserves effective features to the maximum extent.

[0021] Furthermore, the dual-branch parallel feature processing structure is divided into a main feature extraction branch and a direct feature reuse branch. These two branches process independently and in parallel, without interference. The main feature extraction branch abandons the traditional standard convolution used in the native C2f module and instead uses a depthwise separable convolution structure. The spatial filtering and channel fusion operations of conventional convolution are decomposed into two steps: channel-wise depthwise convolution and 1×1 pointwise convolution. This significantly reduces the number of module parameters and floating-point operations from an operational logic perspective. The total number of parameters is calculated using the following formula: Where K is 3, i.e., a conventional 3×3 convolution kernel is used; the direct-pass feature reuse branch does not have any trainable operation layers, but only transmits the channel-compressed features directly to the end fusion unit without loss or delay through the direct-pass link. The feature direct transmission formula is: While preserving shallow detailed features, it avoids the computational waste caused by repeated feature extraction and compensates for the accuracy loss during the lightweight improvement process. After the two branches complete feature processing, the output features are spliced ​​and fused along the channel dimension. The splicing operation formula is as follows: The spliced ​​fused features are fed into the final 1×1 pointwise convolutional feature fusion output layer to complete channel dimension regularization and feature depth integration, and finally output a feature map consistent with the specifications of the native C2f module.

[0022] Furthermore, the core control module is equipped with a 128GB local data storage unit, supporting the storage of raw detection images, videos, and analysis results data categorized by date to avoid data loss. Related data can be exported via external data transmission interfaces such as Gigabit Ethernet and USB 3.0, supporting common formats such as .TIFF / .PNG, .MP4 / .AVI, and .DWG (report). Through pre-stored algorithm models, the terminal can achieve the following core detection functions, with an accuracy rate of no less than 80% for each detection:

[0023] Rust level determination: It is divided into four levels: no rust, light rust, moderate rust, and heavy rust. The rust area is divided by pixel-level segmentation and the rust area percentage is calculated.

[0024] Coating defect identification: Identify eight typical coating defects, including inclusions, exposed substrate, runs, orange peel, pinholes, blistering, rust, and peeling.

[0025] Coating adhesion determination: According to the grid method of SL105-2007 "Code for Corrosion Protection of Hydraulic Metal Structures", it is divided into two categories: qualified and unqualified.

[0026] Surface cleanliness assessment: divided into three categories: clean, relatively clean, and unclean;

[0027] Surface roughness is classified into three categories: fine, medium, and coarse, and is compatible with sandblasting rust removal Sa2.5 and manual rust removal St3 standards.

[0028] Furthermore, the power supply module adopts an 8000mAh rechargeable lithium-ion battery pack, equipped with a wide voltage adaptive charging management chip, supporting a wide range of 9~24V charging, compatible with fast charging and charging-while-using modes, and can support continuous operation for no less than 4 hours under maximum power consumption, ensuring long-term continuous testing on site; the terminal is a non-contact visual inspection structure with an effective detection distance of 0.1m~5m and an optimal detection distance of 0.3m~2m. It can directly control the start and stop of the inspection task, data re-acquisition, and mode switching through the human-machine interaction touch component, without complicated operation, and is suitable for on-site practical operating conditions.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] Integrated design adapts to complex working conditions: The terminal integrates core main control, image acquisition, supplementary lighting, interaction, power supply and other functional components into a one-handed protective carrier. The overall size is 245mm*309mm*78mm and the weight is 1.1kg, making it highly portable. The protection level reaches IP56. With the sealed interface and active heat dissipation design, it can stably adapt to the complex working environment of water conservancy sites with humidity, dust and large temperature fluctuations, and meet the detection needs of special scenarios such as high altitude and confined space.

[0031] Multi-task integrated inspection with precise algorithm adaptation: The terminal uses a pre-stored differentiated and lightweight visual inspection model to complete five core inspection tasks in one go: rust level (4-class segmentation), coating defects (8-class target detection), coating adhesion (2-class classification), cleanliness (3-class classification), and roughness (3-class classification). The algorithm type is precisely matched with the inspection requirements, eliminating the need to carry multiple devices and greatly simplifying the inspection process. A single operator can complete the inspection of a 20-square-meter area in no more than 30 minutes, significantly improving on-site inspection efficiency and reducing manual labor intensity.

[0032] A novel lightweight feature extraction module, through the coupled design of a depthwise separable convolution lightweight structure and a direct feature reuse mechanism, simultaneously achieves a dual reduction in module parameter quantity and computational load while effectively preserving detection accuracy. The module is highly compatible and easy to deploy, requiring no adjustment to the overall structure of the YOLOv8 model, and has no custom operators or non-standard operations. It can be perfectly adapted to various resource-constrained hardware such as ARM architecture embedded devices and mobile terminals, greatly expanding the edge application scenarios of intelligent detection algorithms.

[0033] Offline real-time inference on the edge ensures accurate and reliable detection: The core main control module integrates a neural network computing unit to realize local offline analysis of detection data without relying on the network and cloud computing. Detection results can be obtained in real time on site. The detection process is based on visual algorithms for automatic judgment, avoiding subjective errors caused by human experience. The accuracy rate of each detection is no less than 80%, and the original image and video data can be retained for subsequent traceability and archiving.

[0034] The human-computer interaction is convenient and the operation threshold is low: It is equipped with an industrial-grade anti-interference touch screen and a visual interactive interface, which supports one-stop operation such as parameter configuration, mode switching and result viewing. No professional training is required to get started quickly; the terminal supports convenient control such as starting and stopping detection tasks, data re-acquisition and mode switching, adapting to on-site operating habits and improving the operating experience.

[0035] Stable battery life and flexible data transmission: It adopts a large-capacity rechargeable battery pack, supports fast charging and charging while in use, and meets the needs of long-term continuous operation; it is equipped with wireless and wired dual communication units, which makes data transmission flexible and convenient, and can realize real-time sharing and high-speed export of detection data. Attached Figure Description

[0036] Figure 1 This is a three-dimensional schematic diagram of the overall structure of an embodiment of the portable handheld terminal for detecting the corrosion-resistant surface quality of hydraulic metal structures according to the present invention;

[0037] Figure 2 This is an internal module connection diagram of an embodiment of the portable handheld terminal for detecting the corrosion-resistant surface quality of hydraulic metal structures according to the present invention;

[0038] Figure 3 This is a schematic diagram of the lightweight feature extraction module in one embodiment of the present invention;

[0039] Figure 4 This is a schematic flowchart of one embodiment of the method for detecting the surface quality of corrosion-resistant metal structures in hydraulic engineering according to the present invention.

[0040] Figure 5 This is a picture of a handheld terminal.

[0041] Explanation of reference numerals in the attached diagram: 1. Human-computer interaction touch component; 2. Image acquisition and recognition component; 3. On-site supplementary lighting component; 4. Distance detection unit; 5. Power switch; 6. Screen driver board; 7. Active heat dissipation unit; 8. Core main control module; 9. Power supply unit. Detailed Implementation

[0042] The present invention will be further described in detail below with reference to preferred embodiments. The following embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Equivalent substitutions, conventional parameter adjustments and conventional technical improvements that can be achieved by those skilled in the art without creative effort without departing from the core inventive concept of the present invention all fall within the scope of protection of the present invention.

[0043] Addressing numerous pain points in existing methods for inspecting the corrosion resistance of hydraulic metal structures: traditional inspections rely heavily on manual visual observation, resulting in high subjectivity, low accuracy, and a lack of quantifiable data traceability; conventional inspection equipment is bulky and poorly portable, making it unsuitable for the harsh working environments of hydraulic engineering sites, such as damp, dusty conditions and confined spaces at heights; existing intelligent inspection equipment largely depends on cloud computing power, cannot operate offline, suffers from high data transmission latency, and cannot achieve integrated simultaneous inspection of multiple indicators such as rust level, coating defects, adhesion, cleanliness, and roughness, leading to low operational efficiency. This invention addresses these common industry challenges by proposing an integrated portable intelligent inspection terminal and supporting inspection methods. Through hardware architecture innovation, edge-side intelligent algorithm optimization, and a design specifically adapted for hydraulic engineering scenarios, it achieves non-contact, high-precision, offline simultaneous inspection of multiple indicators, completely resolving the shortcomings of existing technologies.

[0044] Example 1

[0045] like Figure 1 ,5 As shown, this embodiment is a preferred implementation of the core product of the present invention. Through three core innovative designs—multi-module collaborative innovation integration, dedicated optimization of heterogeneous computing power on the end side, and customized adaptation to harsh hydraulic engineering environments—it creates a single-handed, integrated intelligent handheld testing terminal. This breaks through the scene limitations and functional shortcomings of traditional testing equipment, and realizes integrated, portable, and intelligent offline testing of multiple indicators on the anti-corrosion surface of hydraulic metal structures. The entire process does not rely on cloud networks, and the test data is quantified, visualized, and traceable in real time.

[0046] The hardware structure of the integrated portable intelligent inspection terminal includes a human-machine interaction touch control component 1, an image acquisition and recognition component 2, an on-site supplementary lighting component 3, a distance detection unit 4, a power switch 5, a screen driver board 6, an active heat dissipation unit 7, a core main control module 8, and a power supply unit 9. The core of the integrated portable intelligent inspection terminal consists of five main parts: an integrated handheld protective carrier, a heterogeneous computing core processing module, a coaxial image acquisition and adaptive supplementary lighting system, an industrial-grade human-machine interaction module, and a wide-voltage long-life power supply module. Each module adopts a compact integrated architecture, achieving electrical connection and coordinated scheduling through internal customized circuits. This eliminates the drawbacks of the traditional split design. The core processing module acts as the global scheduling and control core, coordinating the orderly operation of each module and achieving parallel processing and synchronous output of multiple inspection indicators.

[0047] like Figure 2 As shown, the internal functional modules of the terminal are electrically connected and scheduled. The core main control module is in the central control position and is directly connected to the image acquisition and recognition component, the on-site lighting component, the human-machine interaction touch component, and the power supply module. The screen driver board connects the touch component and the core main control module. The active heat dissipation unit is controlled by the core main control module. The power supply module supplies power to the entire system. The whole adopts a centralized scheduling architecture. All modules cooperate to achieve integrated control and data transmission. There are no independent redundant links. The structure is simple and the operation is stable.

[0048] The integrated handheld protective carrier is as follows:

[0049] Addressing the unique challenges of humid, dusty, and easily bumped environments at hydraulic engineering sites, and the specific requirements of testing scenarios involving heights, confined spaces, and single-handed operation, this invention abandons the bulky casing and ordinary protective design of traditional testing equipment. Instead, it innovatively employs a composite molding process using high-strength composite engineering materials and a lightweight metal reinforced frame. The overall design features a compact, ergonomic structure designed for single-handed operation, with dimensions strictly adapted to adult single-handed handling habits. The overall weight has been optimized and controlled to 1.1kg, allowing for easy storage in a standard testing toolbox, significantly improving on-site portability and transport convenience.

[0050] In terms of protection, it adopts a fully sealed, waterproof, and dustproof integrated design. Custom-designed sealing strips and compression structures are used at the shell seams, and all external interfaces are equipped with anti-loosening, anti-mud, and dustproof covers. The overall protection level reaches IP56, enabling stable operation in humid, dusty, and slightly corrosive environments at hydraulic engineering sites for extended periods. This solves the problems of traditional equipment being easily damaged and having a high failure rate in harsh environments. Addressing the technical pain point of heat generation during high-computing-power module operation, it innovatively reserves directional heat dissipation channels internally. Combined with high thermal conductivity silicone components and miniature silent active heat dissipation elements, it forms a dual heat dissipation architecture of passive heat conduction and active heat dissipation. This can stably control the operating temperature of core electronic components below 60℃, ensuring that the equipment can withstand long-term high-load offline testing without downtime or performance degradation.

[0051] The carrier's sidewalls are reserved with multi-specification universal installation interfaces, which can be quickly adapted to fixed testing brackets, high-altitude telescopic poles and other supporting tools. It supports multi-angle, long-distance non-contact testing, perfectly adapting to special testing scenarios such as high-altitude gates and narrow corridors, expanding the equipment's application range and filling the technical gaps in areas that cannot be reached by manual testing.

[0052] The core processing module is as follows:

[0053] Addressing the industry's technical bottlenecks of insufficient computing power, reliance on cloud computing, inability to work offline, and inability to perform parallel inference of multiple models in existing detection equipment, this invention innovatively adopts a high-performance edge computing embedded processor with an AI heterogeneous computing architecture. Breaking the computing power limitations of traditional single-core processors, it integrates three major computing power modules: a general-purpose computing core, a graphics acceleration unit, and a dedicated neural network computing power unit, with an overall computing power of 6 TOPS. Coupled with 16GB of high-speed running memory and 128GB of large-capacity local storage, it provides sufficient computing power and storage support for edge-side offline multi-model parallel inference, enabling real-time local processing of detection data and completely eliminating dependence on cloud networks and external computing devices.

[0054] At the system level, the embedded Linux operating system is deeply customized and optimized to meet the requirements of embedded hardware resources and low power consumption. Redundant background processes are eliminated, reducing system power consumption and operating load. The entire initialization process, including device power-on hardware self-test, driver loading, and model preloading, is completed within 20 seconds, enabling rapid boot-up. At the communication level, a multi-mode compatible communication unit is integrated, including high-speed Wi-Fi, Bluetooth, and gigabit Ethernet wired communication modules. This accommodates both convenient wireless data transmission and high-speed wired batch data export, adapting to data transmission requirements in different scenarios. Detection data can be stored locally or uploaded in real time.

[0055] The algorithm is one of the core innovations of this invention: the module pre-installs five types of lightweight, specially optimized visual detection models, covering rust level segmentation, coating defect target detection, coating adhesion classification, cleanliness classification, and roughness classification. All models completely replace the original C2f modules in the YOLOv8 model's backbone feature extraction network and the neck PAN structure through the lightweight feature extraction module. Figure 3 As shown, the module input feature map Size uniformity is characterized as Where H is the feature map height and W is the feature map width. The YOLOv8 model is adaptively adapted based on the number of feature channels at different network layers to ensure compatibility when replacing each layer.

[0056] The pre-channel compression layer uses a 1×1 pointwise convolutional structure with a stride of 1 and padding of 0. The processing does not change the feature map spatial size; it only compresses the number of input feature channels to half the original number. The calculation formula is as follows: Linear activation functions are selected to prevent nonlinear loss of feature information and retain core effective features;

[0057] The dual-branch parallel processing part employs depthwise separable convolution for the main feature extraction branch. The first step performs a 3×3 channel-wise depthwise convolution, independently performing spatial filtering on each channel feature. The parameter calculation formula is as follows: The second step involves performing a 1×1 pointwise convolution to fuse multi-channel feature information. The formula for calculating the number of parameters is as follows: The total number of parameters in the branch is The direct-access feature reuse branch has no trainable parameters or computational layers; it directly applies... Lossless direct transmission to the end of the module, the characteristic direct transmission relationship is as follows: ;

[0058] The feature concatenation and output section will output the main feature extraction branch. With direct feature multiplexing branch output The calculation formula is as follows: (The formula is missing from the original text.) After concatenation, the features are regularized by 1×1 pointwise convolution with a stride of 1. The final output feature map size and number of channels are exactly the same as the original C2f module, ensuring normal and stable operation of subsequent network layers.

[0059] Complete module operation process:

[0060] Receive input feature map transmitted from upper layer network The data is then fed into a pre-convolutional 1×1 channel compression layer to perform channel dimension compression and output the compressed features. ;

[0061] Compressed features The transmission is divided into two parallel paths. One path is fed into the depthwise separable convolution main feature extraction branch to complete the extraction of high-level semantic features of the target and output the features. The other path transmits directly through a straight-through feature multiplexing branch, outputting features. ;

[0062] The output features of the two branches are concatenated along the channel dimension to obtain the fused features. ;

[0063] Fusion features The feature map is fed into the final 1×1 pointwise convolutional layer to complete the feature depth integration and channel dimension regularization, and outputs the final feature map that is compatible with the native C2f module. The map is then transmitted to the next layer network for further computation.

[0064] The module operates without any custom operators or non-standard computational logic. It is compatible with mainstream edge inference frameworks such as ONNX, TensorRT, and NCNN, and can be directly deployed to ARM-based embedded devices, embedded NPUs, and other hardware platforms. According to actual tests, the module reduces the number of parameters by 60% and the floating-point operation by 55% compared to the native C2f module. The accuracy loss of target detection mAP@0.5 is controlled within 2%, balancing lightweight performance and detection accuracy.

[0065] The image acquisition and illumination system is as follows:

[0066] To address the issues of smooth, reflective metal surfaces in hydraulic engineering projects, complex lighting conditions, and the resulting blurry images, lack of detail, and susceptibility to light and shadow interference from traditional acquisition equipment, this invention innovatively adopts an integrated structure where the image acquisition module and the supplementary lighting component are arranged on the same side and coaxially. This ensures that the supplementary lighting beam is coaxial and concentric with the acquisition lens, providing uniform and full coverage of the acquisition area. This completely solves the problems of overexposure due to reflections on metal surfaces, shadow blind spots, and image distortion caused by traditional side supplementary lighting.

[0067] The image acquisition module utilizes an industrial-grade high-definition vision acquisition unit, equipped with automatic focus, automatic exposure, and automatic white balance—three automatic adjustment functions. This allows for precise capture of micron-level defect morphology and true color information on metal surfaces, ensuring the quality of raw data for subsequent model inference. The module connects directly to the core processing module via a high-speed data transmission interface, enabling real-time, zero-latency image data transmission. It also supports automatic adjustment of gain, exposure time, and white balance parameters to adapt to imaging needs of different lighting conditions and metal surfaces of varying materials.

[0068] The supplementary lighting component adopts a high-density, high-power LED light source array with a rated power supply voltage of 24V. It innovatively features a dual-mode control logic of ambient light sensing adaptive adjustment and manual multi-level adjustment: In automatic mode, the built-in light sensor monitors the ambient light intensity in real time and automatically matches the output supplementary lighting brightness to avoid imaging interference caused by reflections from metal surfaces or insufficient lighting; in manual mode, it supports fine adjustment of multiple brightness levels to adapt to extreme lighting scenarios such as dark corridors and strong outdoor light, ensuring stable image quality in all aspects.

[0069] Human-computer interaction, specifically:

[0070] It adopts a 5-inch industrial-grade anti-interference high-brightness touch display with a response time of no more than 1ms, supports multi-touch, and has the characteristics of strong light visibility, anti-static and anti-electromagnetic interference. It can clearly display the picture in direct sunlight outdoors and in dark environments, solving the problems of traditional displays being difficult to see outdoors and lagging operation.

[0071] The user interface features a customized, partitioned visual layout, including a device status bar, a detection mode selection area, a real-time acquisition preview area, a parameter configuration area, and a quick function control area. The operation logic is simplified, supporting one-click switching of detection modes, one-click adjustment of acquisition parameters, real-time preview of the video, and real-time visualization of detection results. No professional training is required for quick and easy operation. Detection results are displayed using a combination of defect annotation boxes, quantitative data, and differentiated color grading. Different defect types and different corrosion / roughness levels are distinguished by unique colors. Clicking on an annotation box retrieves detailed parameters such as confidence level, coordinates, and area percentage. Operators can intuitively and quickly obtain detection conclusions, avoiding subjective errors associated with manual interpretation.

[0072] The power supply module is as follows:

[0073] To address the need for extended operation at hydraulic engineering sites where there is no fixed power source and where continuous operation is required for extended periods, a high-energy-density rechargeable lithium-ion battery pack with a capacity of 8000mAh is adopted. Paired with a dedicated wide-voltage charging management chip, it supports both fast charging and simultaneous charging and use modes. Under maximum power consumption conditions, the device can operate continuously for no less than 4 hours, meeting the needs of long-term continuous testing at the site on a single day.

[0074] The battery pack has a built-in integrated protection board with four safety protection functions: overcharge, over-discharge, overcurrent, and short circuit. It is detachably fixed to the dedicated battery compartment inside the carrier by elastic clips, making replacement convenient and tool-free. The output end uses a DC-DC wide voltage step-down module to convert the voltage into a stable power supply voltage suitable for each module, achieving stable global power supply. At the same time, it is equipped with a real-time power supply status monitoring unit, which automatically cuts off the power in case of power supply abnormality, ensuring the safety of equipment and operators.

[0075] Example 2

[0076] This embodiment, based on the intelligent handheld testing terminal described in Embodiment 1, provides a non-contact, offline, multi-index synchronous surface quality testing method for anti-corrosion of hydraulic metal structures. The entire process is simple to operate, highly efficient, and provides accurate data, without relying on professional testing personnel or external equipment. Figure 4 As shown, the entire testing method follows a top-down linear process, consisting of: equipment startup initialization and hardware self-test; configuration of testing parameters and working modes; coaxial optical path non-contact image acquisition; image preprocessing, noise reduction, and calibration; lightweight YOLOv8 model offline inference on the edge; simultaneous analysis and judgment of five testing indicators; visualization and output of testing results; local classification, storage, and export of data; the entire process requires no cloud network support, with coherent and standardized steps that align with the practical logic of hydraulic engineering operations, achieving full automation of the testing process.

[0077] The specific implementation steps are as follows:

[0078] Step 1: Device power-on initialization, specifically:

[0079] The operator presses and holds the terminal power switch, and the core processing module starts running, automatically completing system loading, full hardware self-test, driver adaptation, and preloading of multiple detection models. The entire initialization process is automated and requires no manual intervention. After initialization, the human-machine interaction module automatically jumps to the main operation interface, and the status bar displays information such as device battery level, communication status, model loading status, and computing power operation status in real time. After confirming that there are no abnormalities in the device, the detection preparation stage can be entered.

[0080] Step 2: Detection mode and parameter configuration, specifically:

[0081] Operators can use a touch interface to select single or multiple target detection modes based on actual on-site testing needs. Upon receiving the command, the core processing module automatically loads the corresponding lightweight detection model. After loading, the interface provides a clear prompt, supporting simultaneous parallel detection of multiple indicators. Two parameter configuration modes are provided to address on-site lighting conditions: First, an automatic adaptation mode, where the terminal, through the image acquisition module and light sensor, monitors ambient light intensity in real time and automatically adjusts exposure parameters, gain, and supplementary lighting brightness without manual intervention; second, a manual fine-tuning mode, where operators can manually fine-tune various parameters based on the actual imaging effect to ensure uniform brightness and clear defect details in the real-time preview image, meeting the image quality requirements for subsequent model inference.

[0082] Step 3: Image acquisition operation, specifically:

[0083] Operators can hold the terminal directly with one hand for testing, or fix the terminal in a suitable position using a bracket or telescopic rod, and align the image acquisition module with the metal surface to be tested. This invention sets the optimal detection distance to 0.3m-2m, with an effective detection distance coverage of 0.1m-5m. Operators can flexibly adjust this distance according to the available space to ensure the entire area to be tested is within the acquisition field of view. After confirming the acquisition position, clicking the "Acquire" shortcut button on the interface initiates the continuous acquisition of multiple high-definition images at a preset frame rate. The supplementary lighting component maintains a constant, adaptive brightness. The acquired raw image data is transmitted in real-time to the core processing module via a high-speed interface, providing a high-quality data source for subsequent offline analysis. The entire process is non-contact, without damaging the metal surface or the original anti-corrosion coating.

[0084] Step 4: Offline intelligent analysis on the device side, specifically:

[0085] After receiving image data, the core processing module first automatically removes blurry, distorted, and overexposed invalid frames using a built-in image filtering algorithm, retaining high-quality and clear images. Then, it calls pre-loaded multi-class detection models and uses heterogeneous computing power to achieve local offline parallel inference analysis, which can be completed independently on the edge without network connection or external computer connection: pixel-level region segmentation, level determination, and area ratio calculation are completed through the corrosion level segmentation model; defect category, location, and confidence level identification are completed through the coating defect detection model; and corresponding index level determination is completed through the adhesion, cleanliness, and roughness classification models.

[0086] Thanks to the heterogeneous computing power optimization and lightweight model design of this invention, the comprehensive inference time for a single frame image is no more than 10 seconds, and the detection efficiency is far higher than that of traditional manual detection and conventional intelligent detection equipment. After the analysis is completed, a complete detection result data package containing all indicators is automatically generated, and the data is complete and traceable.

[0087] Step 5: Result display, storage, and export, specifically as follows:

[0088] The test results are displayed in real-time through a human-computer interaction interface, showing all information including the level of each indicator, pass rate statistics, defect distribution annotations, and core parameter values. The differentiated color-coding is clear and intuitive. When the operator clicks the "Save" button, the terminal automatically stores the original acquired images, annotated images, test result data, and parameter information to a local high-capacity storage medium according to standardized naming rules. It supports TIFF and PNG image formats, MP4 and AVI video formats, and standardized test report formats. For data archiving, external analysis, or reporting, the stored data can be exported at high speed to external devices such as USB flash drives and computers via USB or Gigabit Ethernet interfaces, ensuring stable and efficient data transmission.

[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, component splitting or combination, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A portable handheld terminal for inspecting the corrosion resistance surface quality of hydraulic metal structures, characterized in that, The handheld terminal is a non-contact visual inspection structure, including an integrated handheld protective carrier, a core main control module, an image acquisition and recognition component, an on-site supplementary lighting component, a human-computer interaction touch component, and a power supply module; The core main control module, image acquisition and recognition component, on-site supplementary lighting component, human-computer interaction touch control component and power supply module are all integrated and installed inside the integrated handheld protective carrier. Each functional component is electrically connected to the core main control module and is uniformly scheduled and controlled by the core main control module. The core control module is an embedded intelligent processing unit, equipped with a Linux operating system, an onboard multi-core processor, a graphics processing unit, and a neural network computing unit. It has a locally pre-installed lightweight and optimized visual intelligent detection algorithm model. Through this model, it realizes offline real-time inference calculation on the edge side, receives anti-corrosion surface image data transmitted by the image acquisition and recognition component, and completes five core detection tasks: rust level determination, coating defect identification, coating adhesion determination, rust removal cleanliness and roughness visual evaluation. The detection results are then synchronized to the human-computer interaction touch component for visualization display.

2. The portable handheld terminal for inspecting the corrosion-resistant surface quality of hydraulic metal structures according to claim 1, characterized in that, The integrated handheld protective carrier is an ergonomic shell structure that can be held with one hand. The overall protection level reaches IP56 dustproof and waterproof. The shell has a reserved active heat dissipation air duct. The data transmission interface and charging interface on the outside of the shell are equipped with waterproof sealing rings and anti-loosening buckles.

3. The portable handheld terminal for inspecting the corrosion-resistant surface quality of hydraulic metal structures according to claim 1, characterized in that, The lightweight optimized visual intelligent detection algorithm model includes a rust level segmentation model, a coating defect target detection model, a coating adhesion classification model, a cleanliness classification model, and a roughness classification model. All models replace the native C2f feature extraction module in the backbone and neck structure of the YOLOv8 model with a lightweight feature extraction module. This lightweight feature extraction module consists of a pre-convolutional channel compression layer, a dual-branch parallel feature processing structure, and a post-convolutional feature fusion output layer connected in series. The pre-convolutional channel compression layer performs linear channel compression without changing the feature map spatial resolution; the calculation formula is as follows: In the formula Input the original feature map into the module. It is a 1×1 pointwise convolution kernel weight matrix. For the corresponding convolutional layer bias term, It is a linear activation function.

4. The portable handheld terminal for inspecting the corrosion-resistant surface quality of hydraulic metal structures according to claim 3, characterized in that, The lightweight feature extraction module features a dual-branch parallel feature processing structure, including a main feature extraction branch and a direct feature reuse branch. The main feature extraction branch employs a 3×3 depthwise separable convolution structure, and its total parameter count is calculated using the following formula: In the formula, K is the convolution kernel size, which is fixed at 3. Input the number of channels. Number of output channels; The direct-pass feature multiplexing branch transmits the channel-compressed features directly through the direct link. The feature direct transmission formula is: The output features of the two branches are concatenated and fused along the channel dimension. The concatenation calculation formula is as follows: It balances the dual requirements of feature extraction completeness and model lightweightness, and avoids feature loss.

5. The portable handheld terminal for inspecting the corrosion-resistant surface quality of hydraulic metal structures according to claim 1, characterized in that, The image acquisition and recognition component is fixedly installed on the front detection side of the integrated handheld protective carrier. It adopts an industrial-grade vision acquisition module, which integrates focus adjustment, image parameter adjustment and white balance correction functions. It is used to acquire high-definition static images and real-time dynamic videos of the anti-corrosion surface, and the acquired data is transmitted to the core main control module in real time.

6. The portable handheld terminal for inspecting the corrosion-resistant surface quality of hydraulic metal structures according to claim 1, characterized in that, The on-site supplementary lighting component is arranged on the same side as the image acquisition and recognition component, and the supplementary lighting optical path is coaxially aligned with the image acquisition optical path. The on-site supplementary lighting component uses an LED light source module with a constant brightness and a rated power supply voltage of 24V. It has dual modes of ambient light sensing adaptive adjustment and manual multi-level adjustment, and can automatically supplement the light according to the on-site lighting conditions.

7. The portable handheld terminal for inspecting the corrosion-resistant surface quality of hydraulic metal structures according to claim 1, characterized in that, The core control module is equipped with a 128GB high-capacity local data storage unit, which stores original images, video clips and algorithm analysis results data according to the detection date and detection point. It also integrates a wireless communication unit and a wired communication unit. The wireless unit includes Wi-Fi and Bluetooth modules, while the wired unit adopts Gigabit Ethernet and USB 3.0 interfaces, which takes into account the needs of on-site wireless data transmission and offline high-speed data export, so as to realize the full retention of detection data and subsequent traceability analysis.

8. The portable handheld terminal for inspecting the corrosion-resistant surface quality of hydraulic metal structures according to claim 1, characterized in that, The human-machine interaction touch component adopts a 5-inch industrial-grade anti-interference touch display screen, which is installed on the front operating area of ​​the integrated handheld protective carrier. It is equipped with a customized visual interactive interface, which is divided into a parameter configuration area, a real-time acquisition screen area, a test result display area, and a function quick control area. The test results are displayed visually using defect annotation boxes, quantitative test data, and differentiated color grading. It supports one-click switching of test modes and quick adjustment of acquisition parameters.

9. The portable handheld terminal for inspecting the corrosion-resistant surface quality of hydraulic metal structures according to claim 1, characterized in that, The five core inspection tasks are completed collaboratively using an optimized, lightweight, and intelligent visual inspection algorithm: based on a segmentation algorithm, four levels of rust are determined: no rust, light rust, moderate rust, and heavy rust; based on a target detection algorithm, eight typical coating defects are identified: inclusions, exposed substrate, runs, orange peel, pinholes, blistering, rust, and peeling; and based on a classification algorithm, compliance and grade determination of coating adhesion, rust removal cleanliness, and surface roughness are completed respectively.

10. The portable handheld terminal for inspecting the corrosion-resistant surface quality of hydraulic metal structures according to claim 1, characterized in that, The power supply module adopts a detachable high-capacity lithium-ion battery pack, equipped with a wide voltage adaptive charging management chip, which supports 9~24V wide-range charging, fast charging and charging while using mode. Under the maximum power consumption condition, it can support the terminal to work continuously for no less than 4 hours. The battery pack has a built-in integrated safety protection board with four protection functions: overcharge, over-discharge, overcurrent, and short circuit, making it suitable for long-term operation scenarios in the field without external power supply.

Images