Frost identification and removal method combining machine vision and ultrasonic waves

Abstract

The invention relates to the technical field of refrigeration heat exchanger surface frost treatment, in particular to a machine vision and ultrasonic wave combined frost identifying and removing method. The method comprises the following steps: S1, deploying CCD / CMOS cameras at key points on the surface of the refrigeration heat exchanger to obtain data; s2, processing the data acquired in the step S1; and S3, cooperatively calculating the frosting thickness and area by adopting a Zhang-Suen skeletonization algorithm and an adaptive threshold segmentation algorithm, and judging the frosting level. According to the frost identifying and removing method combining machine vision and ultrasonic waves, ultrasonic defrosting and intelligent control are combined, the defrosting effect is monitored in real time through machine vision, ultrasonic wave parameters are dynamically adjusted, efficient and accurate defrosting is achieved, and the frost identifying and removing method combining machine vision and ultrasonic waves is achieved through multi-technology fusion. The full-automatic process from intelligent frost recognition and accurate positioning to efficient removal is achieved, the intelligent operation and maintenance level of the refrigeration system is greatly improved, manual intervention is reduced, and the operation reliability and stability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of frost treatment on the surface of a refrigeration heat exchanger, and in particular to a frost identification and removal method combining machine vision and ultrasonic waves. Background Art

[0002] There are many pressing challenges in treating frost on the surfaces of refrigeration heat exchangers. First, existing technologies, whether indirect detection based on environmental parameters or manual visual observation, are unable to accurately identify and quantify frost on refrigeration heat exchanger surfaces. Traditional research focuses solely on the impact of environmental conditions on frost thickness and density, lacking precise perception of the frost layer's real-time state, making it difficult to accurately control frost conditions in practical applications. This patent leverages machine vision technology, combined with deep learning algorithms, to automatically learn the morphology, texture, and edge features of frost, outputting a probability map of frost thickness and distribution, enabling accurate identification and quantitative analysis of frost, effectively addressing the issue of inaccurate identification. While widely used, traditional thermal defrosting technology is susceptible to damage from high and low pressure shocks and thermal shocks, and residual droplets on the fin surface after defrosting can easily lead to rapid re-frosting of the evaporator, impacting refrigeration system efficiency and increasing energy consumption. Ultrasonic defrosting technology, while offering advantages, is currently in the research phase, with insufficient understanding of key issues such as defrosting mechanisms and energy consumption, making efficient and energy-saving defrosting difficult to achieve. Existing defrosting technologies rely on preset rules or manual operations and cannot be adaptively adjusted according to actual frost conditions.

[0003] To this end, a frost identification and removal method combining machine vision and ultrasound is designed to provide a technical solution to the above technical problems. Summary of the Invention

[0004] Based on this, it is necessary to provide a frost identification and removal method that combines machine vision and ultrasound to solve the technical problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] A frost recognition and removal method that combines machine vision and ultrasonic waves has the following steps:

[0007] S1: Deploy CCD / CMOS cameras at key points on the surface of the refrigeration heat exchanger to acquire data;

[0008] S2: Process the data collected in step S1;

[0009] S3: The Zhang-Suen skeletonization algorithm and the adaptive threshold segmentation algorithm are used to collaboratively calculate the frost thickness and area, and determine the frost level;

[0010] S4: Match ultrasonic parameters according to frost level.

[0011] As a preferred implementation of the frost identification and removal method combining machine vision and ultrasound provided by the present invention, step S1 also includes synchronously configuring temperature and humidity sensors and infrared thermometers to build a multimodal data acquisition system to obtain a composite data set containing visual information and environmental parameters.

[0012] As a preferred embodiment of the frost identification and removal method combining machine vision and ultrasound provided by the present invention, in step S1, the steps are as follows:

[0013] Construct a dataset to collect images of frost areas on the surface of refrigeration heat exchangers under different operating conditions and frost levels to form a frost dataset on the surface of refrigeration heat exchangers;

[0014] The color, texture and shape features of the image are extracted to build a frost feature database on the surface of the refrigeration heat exchanger.

[0015] As a preferred embodiment of the frost identification and removal method combining machine vision and ultrasound provided by the present invention, in step S2, the steps are as follows:

[0016] The original image is divided into N sub-regions. The OTSU algorithm is used to automatically calculate the optimal threshold value of each sub-region and complete the binarization process to adapt to frost segmentation under different lighting conditions.

[0017] The binary image is expanded and eroded using a 3×3 rectangular kernel structure element to fill frost holes and repair broken edges.

[0018] The Canny algorithm is used to extract the frost edge.

[0019] As a preferred embodiment of the frost identification and removal method combining machine vision and ultrasound provided by the present invention, in step S3, the steps are as follows:

[0020] The marked frost area is iteratively eroded and thinned using the Zhang-Suen skeletonization algorithm, simplifying the frost structure into a single-pixel-wide centerline. This is used as a benchmark to calculate the number of frost skeleton pixels.

[0021] Combined with the total number of pixels in the frost area, the average frost thickness is calculated using the empirical coefficient;

[0022] Adaptive threshold segmentation algorithm is used to automatically calculate the optimal threshold based on the local grayscale distribution characteristics of the image using the Otsu method to accurately separate the frost area from the background.

[0023] By counting the number of frost pixels in the segmented binary image and combining it with the camera calibration parameters, the pixel area is converted into the actual physical area.

[0024] Determine the frost level based on the calculated physical area.

[0025] As a preferred embodiment of the frost identification and removal method combining machine vision and ultrasound provided by the present invention, the frost level is determined based on the calculated physical area, and the steps are as follows:

[0026] Based on preset thresholds, frost levels are divided into light, moderate and heavy;

[0027] Mild refers to thickness less than 1mm and area 10%-30%;

[0028] The moderate thickness is 1-3 mm and the area is 31%-60%;

[0029] The severity is defined as thickness > 3 mm and area > 60%;

[0030] When any sub-region is severely frosted, the global situation is judged as severe.

[0031] As a preferred embodiment of the frost identification and removal method combining machine vision and ultrasound provided by the present invention, in step S4, the steps are as follows:

[0032] If the frost level is light, start the 20kHz / 30W pulse mode, work for 2 minutes, and rest for 1 minute;

[0033] If the frost level is moderate, start the 25kHz / 45W pulse mode and continue working for 4 minutes;

[0034] If the frost level is severe, the 30kHz / 60W pulse mode is activated, and the phase difference is adjusted through the phased array transducer array to achieve energy focusing, giving priority to breaking local thick frost;

[0035] Machine vision is used to monitor the residual frost layer in real time and determine whether the threshold has been reached. If not, the power is automatically increased or a secondary defrost is triggered. During the water melting stage, the system switches to a low-power 15kHz / 20W auxiliary drainage mode based on temperature sensor data to prevent recondensation of liquid water.

[0036] As a preferred embodiment of the frost identification and removal method combining machine vision and ultrasound provided by the present invention, it also includes a multimodal data acquisition module, which is composed of a low-temperature resistant camera, a temperature and humidity sensor, and an infrared thermometer, and is responsible for acquiring images of the heat exchanger surface and environmental parameters;

[0037] Image processing and analysis module, used to perform image preprocessing, multi-algorithm fusion recognition, quantitative analysis and model iterative optimization steps;

[0038] Ultrasonic control module, used to automatically match ultrasonic parameters according to the frost level and drive the transducer array to perform defrost operations;

[0039] The decision-making and early warning module is used to issue graded early warnings based on the identification results, generate maintenance recommendations, and notify operation and maintenance personnel through various means;

[0040] The data management module is used to store historical detection data, model parameters, and equipment status information, and supports data query and analysis.

[0041] It can be seen without a doubt that the above-mentioned technical solution of this application can definitely solve the technical problem to be solved by this application.

[0042] At the same time, through the above technical solutions, the present invention has at least the following beneficial effects:

[0043] 1. The present invention provides a frost identification and removal method that combines machine vision and ultrasound. It combines ultrasonic defrosting with intelligent control, monitors the defrosting effect in real time through machine vision, and dynamically adjusts ultrasonic parameters to achieve efficient and accurate defrosting. Compared with traditional thermal defrosting, energy consumption is greatly reduced, while equipment damage and heavy frosting problems are avoided. Through the integration of multiple technologies, a fully automated process from intelligent frost identification and precise positioning to efficient removal is realized, which greatly improves the intelligent operation and maintenance level of the refrigeration system, reduces manual intervention, and improves the reliability and stability of operation.

[0044] 2. The present invention combines the precise recognition capability of machine vision with the high efficiency of ultrasonic defrosting to achieve intelligent recognition and dynamic processing of frost on the surface of the refrigeration heat exchanger. At the same time, it has the characteristics of high adaptability, low energy consumption, equipment friendliness, and full-process automation, which significantly improves the operating efficiency and reliability of the refrigeration heat exchange system, thereby solving the problems existing in the prior art such as inaccurate frost recognition, low defrosting efficiency, fragile equipment, high energy consumption, and lack of intelligent control.

[0045] 3. The present invention establishes a complete automated processing and intelligent control process, starting with the camera automatically capturing images of the refrigeration heat exchanger surface, and proceeding sequentially through image preprocessing, frost feature extraction and recognition, thickness and area calculation, frost level determination, and finally automatically matching the ultrasonic defrost strategy and real-time monitoring of the defrost effect. The entire process requires no human intervention. The system uses an optimized algorithm architecture and parallel computing technology to quickly process image data and control instructions, achieving real-time monitoring and response to frost. Existing technologies often rely on fixed parameter defrosting or manual judgment to initiate defrosting, which is not only inefficient but also prone to equipment loss and energy waste. After achieving full process automation, the present invention supports all-weather real-time monitoring and can dynamically adjust the ultrasonic frequency, power, and action time according to the degree of frost. Compared with traditional thermal defrosting methods, the defrost efficiency is improved by more than 40%, energy consumption is reduced by more than 50%, and the high and low pressure shock and thermal shock to the equipment caused by traditional thermal defrosting are avoided, extending the equipment service life by more than 25%.

[0046] 4. This invention combines multi-algorithm fusion of machine vision with ultrasonic dynamic control. For machine vision, an industrial-grade, low-temperature-resistant, high-resolution camera is used to capture images. The original image is divided into N subregions according to a specific ratio. The adaptive Otsu threshold or Gaussian clustering model is independently calculated for each subregion's local grayscale distribution (mean, variance). This facilitates high-precision processing of the target subregion, ignores irrelevant background areas, reduces ineffective calculations, and thus improves calculation speed for frost-contaminated areas.

[0047] 5. The present invention uses the adaptive threshold Otsu method combined with edge detection to automatically calculate the optimal threshold and accurately extract the edge contour of frost on the surface of the refrigeration heat exchanger, overcoming the defect that the traditional fixed threshold method is difficult to adapt to different lighting and changes in frost characteristics; on this basis, the Gaussian mixture model can capture the complex texture, shape, color and other characteristic distributions of frost by probabilistic modeling of pixel distribution, and more accurately distinguish frost from the heat exchanger surface; in actual applications, the traditional fixed threshold segmentation method often suffers from over-segmentation or under-segmentation in scenarios where the heat exchanger surface is reflective and frost is uneven, resulting in large errors in the calculation of frost thickness and area; the combination of the adaptive threshold Otsu method and Canny edge detection of the present invention greatly improves the accuracy of edge contour extraction of heat exchanger surface images with different lighting and frost levels.

[0048] 6. In the image preprocessing stage of the present invention, the color image is converted into a grayscale image through grayscale processing, which reduces the interference of color information and makes the algorithm focus on the grayscale features; denoising operations such as median filtering and Gaussian filtering effectively suppress environmental noise, and the binarized image is optimized in combination with morphological operations (dilation and corrosion), filling the internal gaps of the frost, connecting scattered frost blocks, and eliminating the influence of stains on the surface of the heat exchanger; the operating environment of the refrigeration heat exchanger is complex, and there are interference factors such as sudden temperature changes, humidity fluctuations, and surface oil stains. The existing technology is prone to false detection and missed detection in such an environment. The present invention effectively filters out irrelevant interference factors through multi-step preprocessing, providing a stable input for the core algorithm; in a complex operating environment, the false detection rate and missed detection rate are effectively reduced, which greatly improves the adaptability and reliability of the algorithm in a complex environment compared to the existing technology.

[0049] 7. This invention not only intelligently identifies frost but also achieves refined assessment of frost severity through dual-dimensional quantitative analysis (frost thickness and coverage area ratio) and a three-level precision grading mechanism. It employs a combined determination based on both thickness and area (light frost requires both "thickness <1mm" and "coverage 10%-30%") to avoid misjudgment based on a single criterion. The detection results of each sub-area are integrated into a global classification using the "highest level first" principle (heavy frost in any sub-area will result in a global classification of severe severity), ensuring that localized thick frost issues are not masked. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 This is a schematic diagram of the intelligent recognition of frost on the surface of a refrigeration heat exchanger according to the present invention;

[0052] Figure 2 This is a schematic diagram of the frost coverage area ratio and average thickness results after image segmentation processing according to the present invention;

[0053] Figure 3 This is an overall flow chart of the method for intelligently identifying and efficiently removing frost on the surface of a refrigeration heat exchanger according to the present invention;

[0054] Figure 4 A schematic diagram of intelligent frost recognition and ultrasonic graded defrosting on the surface of a refrigeration heat exchanger according to the present invention;

[0055] Figure 5 Schematic diagram of the hardware system of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be 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 the present invention and are not intended to limit the present invention.

[0057] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0058] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features and technical solutions therein may be combined with each other.

[0059] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not require further definition or explanation in subsequent drawings.

[0060] Example 1

[0061] Reference Figure 1-Figure 5 ,A frost identification and removal method combining machine vision and ultrasonic waves.

[0062] A CCD / CMOS camera is deployed at the key point of the refrigeration heat exchanger surface, and a temperature and humidity sensor and an infrared thermometer are configured synchronously to build a multi-modal data acquisition system and obtain a composite dataset containing visual information and environmental parameters. The collected raw images are divided into N sub-regions (e.g., a 4x4 grid), and each sub-region is sequentially subjected to grayscale conversion, median filtering (5x5 kernel), and adaptive histogram equalization to eliminate noise and enhance contrast, thereby solving the problem of spatial non-uniformity of thin frost at the fin root and thin frost at the pipe edge. OTSU threshold segmentation, morphological operation (2 times of 3x3 structure element dilation + 1 time of erosion), and Canny edge detection (double threshold 50 / 150) are performed on each sub-region, and a pixel is classified by a Gaussian mixture model to accurately extract the frost area. Based on the camera calibration parameters, the average thickness (error ±0.1 mm) is estimated by the Zhang-Suen skeletonization algorithm, the pixel number is converted into the actual area (error <5%), and the frost level is determined based on the "thickness and area" double conditions. The frost level is divided into light, medium, and heavy according to the preset threshold (light: thickness <1 mm and area 10%-30%; medium: thickness 1-3 mm and area 31%-60%; heavy: thickness >3 mm and area >60%), and the global frost level is determined as heavy if any sub-region is heavy. The ultrasonic wave parameters (20 kHz, 30 W low-level cleaning; 25 kHz, 45 W medium-level cleaning; 30 kHz, 60 W high-level cleaning) are automatically matched according to the frost level. 10% of the test samples are periodically extracted for comparison with high-precision laser thickness data, and the GMM clustering parameters and U-Net network structure are optimized using the SHAP value analysis algorithm to improve the recognition accuracy. The system uses a distributed computing architecture, the camera end performs preprocessing, the edge computing node runs the GMM clustering and U-Net model, and the cloud server is responsible for model training, thereby achieving ultra-fast real-time response through three-level collaboration. Blockchain technology is integrated to encrypt and store the test data and defrosting records, ensuring that the data cannot be tampered with. The specific steps are as follows:

[0063] First, image data of key points on the surface of the refrigeration heat exchanger under different working conditions is collected in the field, and an industrial-grade low-temperature-resistant and high-integration CCD / CMOS high-resolution camera (with a protection level of IP65 or above and a resolution of not less than 4K) is installed at key points such as the fin gap and the pipe surface of the refrigeration heat exchanger. The heat exchanger surface in different frosting levels (thin frost, medium frost, and thick frost) is photographed from multiple angles under various complex environmental conditions, and a temperature and humidity sensor and an infrared thermometer are configured synchronously to build a multi-modal data acquisition system and obtain a composite dataset containing visual information and environmental parameters.

[0064] A dataset was constructed by collecting images of frost areas on the surfaces of refrigeration heat exchangers under different operating conditions (temperature -25°C to 10°C, humidity 60% to 95%) and frost levels (light frost, moderate frost, and heavy frost). The images were then analyzed by extracting color features (mean and variance in HSV space), texture features (energy and entropy of the gray-level co-occurrence matrix (GLCM), and shape features (perimeter and circularity) to construct a database of frost features on refrigeration heat exchanger surfaces.

[0065] All collected images are normalized to enhance model accuracy and generalization capabilities, and reduce deviations caused by data scale differences. The normalized complete image is divided into multiple adaptive sub-regions, and pre-processing, feature extraction, and frost analysis are performed independently for each sub-image. This can effectively address the spatial non-uniformity of frost on the surface of the refrigeration heat exchanger (thicker frost at the root of the fin and thinner frost at the edge of the pipe), and avoid feature confusion caused by global processing. Through block detection, the thickness of the frost layer in the local thick frost area can be accurately identified to prevent the overall average calculation from masking the local serious frost problem; at the same time, the smaller sub-image size can reduce the interference of background noise such as stains on the heat exchanger surface, light and shadow reflections on the frost characteristics, significantly improve the accuracy of edge detection and threshold segmentation, and achieve refined analysis and identification of the frost conditions in each area.

[0066] Next, adaptive histogram equalization is applied to the segmented sub-images to enhance image contrast. Median filtering (using a 5×5 filter kernel) is then used to remove salt-and-pepper noise and smooth image edges. Gaussian filtering (using a 5×5 Gaussian kernel) is also applied to further eliminate Gaussian noise. The integrated Python-based image processing algorithm includes functions such as weighted average grayscale conversion, binarization, morphological operations, edge detection, and frost thickness calculation. The sub-images are converted from color images to grayscale using BGR2GRAY to reduce color interference and reduce data processing complexity. OTSU binarization is used to automatically calculate the optimal threshold and separate the image into foreground (frost) and background (heat exchanger surface). Pixels below the threshold are set to white (background) and pixels above the threshold are set to black (potential frost areas). This allows accurate differentiation between frost and the heat exchanger surface in various environmental conditions. Next, morphological operations (dilation and erosion) are performed. A 3×3 kernel is used to perform two dilations to fill small holes in the frost area and connect adjacent frost blocks. The dilated image is then eroded once to remove edge burrs and restore the true boundary of the frost.

[0067] Canny edge detection is performed on the grayscale image, with a low threshold of 50 and a high threshold of 150 used to identify reliable edges. Region fusion is performed using bitwise_or to fuse the morphologically processed binary image (Eroded) with the edge image (Edges), integrating the complete region with the frost edge details to extract frost features from the sub-image.

[0068] Preferably, GMM is combined with probabilistic pixel classification to integrate multi-feature recognition results through a weighted fusion strategy. The pre-processed sub-regions are input into the trained frost intelligent recognition model, and through convolution, pooling, and upsampling operations, the frost region segmentation is further refined to improve recognition accuracy.

[0069] After frost feature extraction, an adaptive threshold model is used to classify image pixels. The EM algorithm is used to iteratively optimize model parameters and learn the distribution of frost feature variations, accurately separating frosted and non-frosted areas and improving recognition accuracy. Combined with adaptive threshold segmentation based on local entropy, the threshold is automatically adjusted based on the information entropy of each 32×32 subregion of the image to adapt to varying lighting conditions and surface material variations of the heat exchanger. To address the uneven lighting and varying surface reflectivity experienced during refrigeration system operation, adaptive threshold segmentation calculates an appropriate threshold for each local region, enabling clearer separation of frosted areas from the background. Based on the automatic recognition basic model, new images of the refrigeration heat exchanger surface are continuously collected as processing images and fed into the model for recognition, resulting in recognition results. A random sample of 10% of the recognition results is then validated and analyzed through manual review or comparison with results from high-precision laser thickness measurement equipment with an accuracy of ±0.05 mm. For erroneous or biased recognition results, the SHAP value analysis algorithm was used to thoroughly analyze the causes, reversely optimize model parameters, and adjust the image feature extraction algorithm and edge detection strategy to eliminate the influence of uncertain factors such as ambient lighting changes and differences in heat exchanger surface materials. Through multiple rounds of iteration, the accuracy of automatic frost recognition results was improved and the reliability of the model was verified, resulting in an optimized intelligent frost recognition model for refrigeration heat exchanger surfaces. Based on this optimized automatic recognition model, the powerful feature extraction capabilities of convolutional neural networks were combined to model the changing characteristics of frost on refrigeration heat exchanger surfaces. Four convolutional and pooling layers were used to extract deep features of frost at different growth stages.

[0070] Finally, for area calculation, the camera is pre-calibrated using precision measuring equipment to obtain pixel density parameters. The actual frost area is calculated using a formula. The percentage of the frost area to the total heat exchanger area is also calculated to determine the severity of frost. Thickness estimation creates a binary mask for the frost area and uses the Zhang-Suen skeletonization algorithm to extract the frost skeleton, simplifying the frost morphological structure. Based on the relationship between the number of skeleton pixels and the number of pixels in the frost area, an empirical coefficient is used to calculate the average thickness in pixels. This is then converted to actual thickness in millimeters using the corresponding formula. Frost thickness and area calculation uses the camera calibration parameters previously obtained through the checkerboard calibration method to convert the number of pixels in the marked area into actual area and thickness. The severity of frost is graded according to the preset three-level warning threshold, including mild warning (frost thickness <1mm and coverage area accounts for 10%-30% of the heat exchanger surface area), moderate warning (frost thickness 1-3mm and coverage area accounts for 31%-60% of the heat exchanger surface area), and high warning (frost thickness >3mm and coverage area accounts for >60% of the heat exchanger surface area), and supports user-defined adjustment. After each sub-area independently completes the frost feature analysis, the system integrates the sub-area results into a global level. If any sub-area is judged to be severely frosted, the global level is directly judged as severe; if there are multiple sub-area levels, the level with the largest proportion is used as the global frost level (if the area of ​​the moderate sub-area accounts for >50%, the global level is judged as moderate). Based on the frost level results, the decision-making and warning module retrieves the corresponding ultrasonic parameters from the built-in parameter mapping table, and sends them to the ultrasonic control module in real time through the edge computing node to drive the transducer array to perform defrosting operations.

[0071] Preferably, the segmented frost image is converted into a binary image, and the number of pixels with a value of 1 (representing frost) is counted to obtain the frost pixel area. For multiple unconnected frost areas, a seed filling algorithm is used to mark each area, and the number of pixels is counted and accumulated. The actual area (unit: cm) is converted based on the camera calibration parameters. 2 ).

[0072] During the defrost process, machine vision monitors frost thickness in real time. If the residual thickness after a single defrost cycle exceeds 0.5mm, the system automatically increases power by one level (from mild to moderate parameters) and triggers a second defrost. Combined with temperature sensor data, the system switches to low-power ultrasonic drainage during the frost melting phase to prevent refreezing of liquid water. If the standard is still not met after two consecutive adjustments, a manual intervention alert is sent via SMS or app notification. Defrosting is divided into an "ice-breaking phase" (high power for rapid defrosting) and a "water-melting phase" (low power for drainage). The ice-breaking phase uses power levels corresponding to the frost level (20kHz / 30W to 30kHz / 60W) and continues until the frost thickness is less than 1mm. During the water-melting phase, the system switches to low power mode (15kHz / 20W) and, combined with temperature sensor data (when the surface temperature is >0°C), maintains this mode for 5-10 minutes to remove residual droplets and prevent recondensation. The system dynamically optimizes parameter combinations by calculating defrost efficiency (Δthickness / Δtime) and energy consumption ratio (frost mass / energy consumption) in real time. If the defrost efficiency is less than 0.3mm / min and the energy consumption ratio is greater than 1.5kJ / g, it will automatically switch to a higher frequency gear to avoid energy waste while ensuring efficiency.

[0073] For specific frost levels and comparisons, please refer to Table 1 and Table 2.

[0074] Table 1 Frost level-ultrasonic parameter mapping table

[0075]

[0076] The built-in parameter mapping table (see Table 1) is automatically called according to the frost level to form a gradient defrost strategy. The 20kHz / 30W pulse mode is started lightly (working for 2 minutes and resting for 1 minute), switched to 25kHz / 45W for 4 minutes in moderate conditions, and 30kHz / 60W is activated in severe conditions and the phase difference is adjusted through the phased array transducer array to achieve energy focusing, giving priority to breaking local thick frost. During the defrost process, machine vision monitors the residual frost layer in real time (threshold 0.5mm). If it does not meet the standard, the power is automatically increased or a second defrost is triggered; in the water melting stage, combined with the temperature sensor data, it switches to a low power of 15kHz / 20W to assist in drainage to prevent liquid water from recondensing.

[0077] Table 2 Comparison of frost level and dynamic matching effect of ultrasonic parameters

[0078]

[0079] In other embodiments, evaluation metrics such as accuracy, recall, and F1 score can be used to verify model reliability, and the contribution of each feature to the recognition result can be quantified using a SHAP value analysis algorithm. The model structure can be optimized by adjusting the number of clusters (K value) in the GMM and the size and number of convolutional kernels in the U-Net based on the error. Data augmentation operations such as rotation, scaling, and brightness adjustment can be performed on the training dataset to improve model generalization.

[0080] Example 2

[0081] This disclosure is based on the above-mentioned embodiment 1.

[0082] Based on the percentage and thickness of frost coverage, the frost level is pushed in real time through indicator lights, APP reminders and SMS notifications, and the frost area, thickness and level information are displayed simultaneously. Maintenance recommendations are automatically generated. For light frost, routine cleaning is recommended within 72 hours; for moderate frost, deep cleaning is recommended within 24 hours; and for heavy frost, the defrost program is immediately triggered and the operation and maintenance personnel are notified.

[0083] Example 3

[0084] A system is disclosed based on the above-mentioned embodiment 1.

[0085] It includes a multimodal data acquisition module, specifically composed of a low-temperature-resistant camera, a temperature and humidity sensor, and an infrared thermometer, responsible for acquiring images of the heat exchanger surface and environmental parameters; an image processing and analysis module, which performs the image preprocessing, multi-algorithm fusion recognition, quantitative analysis, and model iterative optimization steps proposed in Example 1. An ultrasonic control module automatically matches ultrasonic parameters according to the frost level and drives the transducer array to perform defrost operations; a decision-making and early warning module, which provides graded early warnings based on the recognition results, generates maintenance recommendations, and notifies operations and maintenance personnel through various means. A data management module stores historical detection data, model parameters, and equipment status information, and supports data query and analysis.

[0086] Preferably, the system adopts a distributed computing architecture. The camera side performs image preprocessing and feature extraction, the edge computing node runs GMM clustering and U-Net model, and the cloud server is responsible for model training and data management. The three-level collaborative processing mechanism improves the real-time performance and reliability of the system.

[0087] The ultrasonic control module includes a phased array transducer array. By adjusting the phase difference of each transducer, the ultrasonic energy is focused on a specific area, improving the efficiency of thick frost removal.

[0088] Preferably, the system integrates blockchain technology to encrypt and store detection data, defrost records, and maintenance logs to ensure that the data cannot be tampered with, providing a reliable basis for the management of the entire life cycle of the equipment.

[0089] Achieved results:

[0090] 1. It can solve existing problems such as inaccurate frost identification, low defrosting efficiency, fragile equipment, high energy consumption, and lack of intelligent control. By combining the precise recognition capabilities of machine vision with the efficient characteristics of ultrasonic defrosting, this invention achieves intelligent identification and dynamic processing of frost on the surface of refrigeration heat exchangers. It also features high adaptability, low energy consumption, equipment-friendly operation, and full process automation, significantly improving the operating efficiency and reliability of refrigeration and heat exchange systems.

[0091] 2. Through the collaborative calculation of the Zhang-Suen skeletonization algorithm and the OTSU threshold segmentation, the frost layer thickness error is controlled within ±0.1mm, and the area ratio error is less than 5%, which is a significant improvement compared to the existing technology (human visual inspection error >20%). Based on the precise classification results, the system automatically matches the gradient ultrasonic defrost strategy through built-in parameter mapping (see Table 2), achieving dynamic response of low-power pulse defrosting for light frosting and energy focusing enhancement for heavy frosting. For light frosting, a 20kHz / 30W pulse mode (working for 2 minutes and resting for 1 minute) is used, which reduces energy consumption by more than 50% compared with traditional thermal defrosting, avoiding excessive defrosting; for moderate frosting, a 25kHz / 45W continuous working mode is enabled, combined with an adaptive power regulation algorithm (dynamically adjusting the power by ±10% according to the thickness of the frost layer), and defrosting is completed within 4-6 minutes. The energy efficiency ratio (defrosting amount / energy consumption) is greatly improved compared with the fixed power mode, taking into account both efficiency and energy consumption; for heavy frosting, a 30kHz / 60W power mode is enabled and combined with phased array focusing technology, the energy density is increased to twice that of the conventional mode, and the defrosting efficiency is increased by 40% compared with single ultrasonic technology. The secondary defrost is automatically triggered by real-time monitoring of the residual frost layer (threshold 0.5mm) to ensure that no thick frost remains.

[0092] 3. Compared with the existing technology, the present invention breaks through the bottleneck of manual judgment lag and fixed parameter inefficiency. It synchronizes the frost level in real time through indicator lights, APP and text messages, and automatically generates maintenance suggestions (light frost is recommended to be cleaned within 72 hours, moderate frost is recommended to arrange deep cleaning within 24 hours, and heavy frost triggers defrosting immediately), reducing the frequency of manual inspections. Dynamic parameter matching greatly reduces the average energy consumption. Combined with the "ice breaking + melt water" two-stage control, repeated defrosting caused by liquid water recondensation is avoided; phased array focusing technology is used for targeted defrosting of areas prone to thick frost accumulation, such as the root of the fin, to avoid excessive impact of traditional full-area defrosting on non-frosting areas, thereby extending the service life of the equipment. The present invention constructs a closed-loop intelligent control system for frost treatment of refrigeration heat exchangers through full-process automation, which solves the pain points of inaccurate recognition, fuzzy judgment and high energy consumption of traditional technologies, and provides core technical support for the intelligent upgrade of refrigeration systems.

[0093] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A frost identification and removal method combining machine vision and ultrasonic waves, characterized in that: Here are the steps: S1: Deploy CCD / CMOS cameras at key points on the surface of the refrigeration heat exchanger to acquire data; S2: Process the data collected in step S1; S3: The Zhang-Suen skeletonization algorithm and the adaptive threshold segmentation algorithm are used to collaboratively calculate the frost thickness and area, and determine the frost level; S4: Match ultrasonic parameters according to frost level.

2. The frost identification and removal method combining machine vision and ultrasound according to claim 1 is characterized in that: Step S1 also includes synchronously configuring temperature and humidity sensors and infrared thermometers to build a multimodal data acquisition system to obtain a composite data set containing visual information and environmental parameters.

3. The frost identification and removal method combining machine vision and ultrasound according to claim 2 is characterized in that: In step S1, the steps are as follows: Construct a dataset to collect images of frost areas on the surface of refrigeration heat exchangers under different operating conditions and frost levels to form a frost dataset on the surface of refrigeration heat exchangers; The color, texture and shape features of the image are extracted to build a frost feature database on the surface of the refrigeration heat exchanger.

4. The frost identification and removal method combining machine vision and ultrasound according to claim 1 is characterized in that: In step S2, the steps are as follows: The original image is divided into N sub-regions. The OTSU algorithm is used to automatically calculate the optimal threshold value of each sub-region and complete the binarization process to adapt to frost segmentation under different lighting conditions. The binary image is expanded and eroded using a 3×3 rectangular kernel structure element to fill frost holes and repair broken edges. Use the Canny algorithm to extract frost edges.

5. The frost identification and removal method combining machine vision and ultrasonic waves according to claim 1 is characterized in that: In step S3, the steps are as follows: The marked frost area is iteratively eroded and thinned using the Zhang-Suen skeletonization algorithm, simplifying the frost structure into a single-pixel-wide centerline. This is used as a benchmark to calculate the number of frost skeleton pixels. Combined with the total number of pixels in the frost area, the average frost thickness is calculated using the empirical coefficient; Adaptive threshold segmentation algorithm is used to automatically calculate the optimal threshold based on the local grayscale distribution characteristics of the image using the Otsu method to accurately separate the frost area from the background. By counting the number of frost pixels in the segmented binary image and combining it with the camera calibration parameters, the pixel area is converted into the actual physical area. Determine the frost level based on the calculated physical area.

6. The frost identification and removal method combining machine vision and ultrasonic waves according to claim 5 is characterized in that: According to the calculated physical area, determine the frost level as follows: Based on preset thresholds, frost levels are divided into light, moderate and heavy; Mild refers to thickness less than 1mm and area 10%-30%; The moderate thickness is 1-3 mm and the area is 31%-60%; The severity is defined as thickness > 3 mm and area > 60%; When any sub-region is severely frosted, the global situation is judged as severe.

7. The frost identification and removal method combining machine vision and ultrasonic waves according to claim 6 is characterized in that: In step S4, the steps are as follows: If the frost level is light, start the 20kHz / 30W pulse mode, work for 2 minutes, and rest for 1 minute; If the frost level is moderate, start the 25kHz / 45W pulse mode and continue working for 4 minutes; If the frost level is severe, the 30kHz / 60W pulse mode is activated, and the phase difference is adjusted through the phased array transducer array to achieve energy focusing, giving priority to breaking local thick frost; Machine vision is used to monitor the residual frost layer in real time and determine whether the threshold has been reached. If not, the power is automatically increased or a secondary defrost is triggered. During the water melting stage, the system switches to a low-power 15kHz / 20W auxiliary drainage mode based on temperature sensor data to prevent recondensation of liquid water.

8. The frost identification and removal method combining machine vision and ultrasonic waves according to claim 1 is characterized in that: It also includes a multimodal data acquisition module, which consists of a low-temperature resistant camera, temperature and humidity sensors, and an infrared thermometer, responsible for acquiring images of the heat exchanger surface and environmental parameters; Image processing and analysis module, used to perform image preprocessing, multi-algorithm fusion recognition, quantitative analysis and model iterative optimization steps; Ultrasonic control module, used to automatically match ultrasonic parameters according to the frost level and drive the transducer array to perform defrost operations; The decision-making and early warning module is used to issue graded early warnings based on the identification results, generate maintenance recommendations, and notify operation and maintenance personnel; The data management module is used to store historical detection data, model parameters, and equipment status information, and supports data query and analysis.

Images