Clothing cleanliness intelligent evaluation system and method based on cloud computing

Abstract

This invention discloses a cloud-based intelligent system and method for assessing clothing cleanliness, belonging to the field of clothing cleanliness assessment technology. The invention identifies clothing labels and performs intelligent cleaning; controls the clothing to ensure uniformity of the lint; uses a deep learning-based lint density assessment model to evaluate the clothing, issuing a cleanliness assessment command when the uniformity of the lint is determined to be satisfactory; extracts multispectral features from clothing images and uses the YOLOv5 model for stain recognition; calculates the stain area ratio based on the stain recognition results, classifies the stain severity, and calculates a cleanliness score; uses a ray tracing algorithm to simulate the reflection and occlusion of light on the lint surface to obtain the natural shadow color of the lint; converts the stain colors and natural shadow colors of the lint from the stain database to the CIE Lab color space and calculates the color difference; calculates the impact of the color difference on the cleanliness score and adjusts the cleanliness score accordingly.

Description

Technical Field

[0001] This invention relates to the field of clothing cleanliness assessment technology, specifically a cloud-based intelligent clothing cleanliness assessment system and method. Background Technology

[0002] With consumers' increasing demands for clothing cleanliness and the widespread use of plush garments in the textile industry, the importance of clothing cleanliness assessment and cleaning technology is becoming increasingly prominent. Traditional clothing cleaning processes rely heavily on manual experience and lack precision and automation. In recent years, although some cleanliness assessment technologies based on visual recognition and sensor detection, as well as automated cleaning equipment, have been put into application, many technical bottlenecks still exist when handling plush garments.

[0003] Current visual recognition technologies struggle to distinguish the natural shadows of fluffy fibers from stain colors in plush garments, easily mistaking shadows for stains or missing deep stains, leading to inaccurate cleanliness assessments. Furthermore, traditional detection methods are weak at identifying stains in tightly packed fiber areas such as seams, where stains are compressed into gaps, resulting in slight surface color changes that conventional methods cannot effectively detect. The process from garment washing to cleanliness assessment is largely independent, lacking coordination. For example, washing parameters cannot be automatically adjusted based on garment material, and fluff finishing requires manual operation, preventing fully automated processing. For plush garments, existing washing and finishing technologies cannot guarantee even fluff distribution, easily leading to tangling, knots, or localized accumulation, affecting the garment's appearance and texture. Simultaneously, inaccurate cleanliness assessments can result in over- or under-washing, damaging the garment material or leaving stains. Summary of the Invention

[0004] The purpose of this invention is to provide a cloud computing-based intelligent assessment system and method for clothing cleanliness to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Firstly, this application provides a cloud-based intelligent assessment method for clothing cleanliness, including the following steps:

[0007] The system identifies the clothing label, determines that the clothing is made of plush material, retrieves the corresponding washing parameters, and performs intelligent washing. It also performs physical combing control, airflow-assisted control, and temperature setting control on the clothing to ensure that the pile of the clothing is evenly distributed.

[0008] The system collects images of clothing, performs image preprocessing, and uses a deep learning-based fluff density assessment model to evaluate the clothing. When it is determined that the uniformity of the fluff in the clothing meets the standard, a cleanliness assessment command is issued.

[0009] When a cleanliness assessment instruction is received, multispectral features are extracted from the clothing image, and stains are identified using the YOLOv5 model. Based on the stain identification results, the stain area ratio is calculated, the stain severity is graded, and a cleanliness score is calculated.

[0010] The physical properties of the velvet were obtained, and the reflection and occlusion of light on the velvet surface were simulated using a ray tracing algorithm. The simulated shadow image was divided into regions, and the average color value of each region was calculated to obtain the natural shadow color of the velvet. The stain colors and natural shadow colors of the velvet in the stain database were converted to the CIE Lab color space, and the color difference was calculated. The impact of the color difference on the cleanliness score was calculated, and the cleanliness score was adjusted accordingly.

[0011] In conjunction with the first aspect, in the first embodiment of the first aspect of this application, the step of identifying the clothing label, wherein the clothing material is plush, retrieving the corresponding washing parameters, and performing intelligent washing includes:

[0012] The system identifies clothing tags using RFID readers or QR code scanners; extracts the material field to determine if the clothing is made of plush or fleece; if it is, it proceeds to the next step; otherwise, it retrieves the corresponding material cleaning parameters and executes other cleaning processes; it retrieves the cleaning parameters for plush clothing from a preset parameter database and sends start commands to the ultrasonic cleaner and detergent dispenser; during the cleaning process, it monitors the operating status of the cleaning equipment in real time; after the cleaning time is completed, it triggers the rinsing program; it detects the conductivity of the cleaning solution using a conductivity sensor to determine the amount of detergent residue; when the conductivity reaches the preset rinsing end standard, the water quality is considered to be up to standard, the rinsing program is stopped, and the intelligent cleaning process is completed.

[0013] In conjunction with the first aspect, in the second embodiment of the first aspect of this application, the physical combing control, airflow-assisted control, and temperature-setting control of the clothing to achieve uniform fabric pile includes:

[0014] An industrial camera scans the garment, and image recognition technology locates areas with tightly packed fibers, marking the areas that need combing. An electric soft brush is then moved over the garment, and a servo motor adjusts the brush head angle according to the marked areas. A pressure sensor at the brush tip provides real-time feedback on the contact pressure with the garment, adjusting the servo motor torque to maintain a stable combing force. The brush moves at a constant speed, starting from one end of the garment and inserting into the fiber gaps for combing. After combing a fixed distance, an ion air gun is activated to blow away dust and impurities carried out during the combing process, preventing them from re-adhering to the fibers, before continuing to the next section. This process is repeated until all marked areas have been combed.

[0015] After combing, the garments are hung in designated locations. The motorized gimbal automatically adjusts the blower outlet based on the seam angle and turns on the blower. An industrial camera captures real-time images of the fluff's movement under the airflow, analyzing the uniformity of fluff distribution. If uneven fluff distribution is detected, the blower speed is increased; if fluff scatters, the speed is decreased. After the blower continues blowing for a period, it is paused, and the fluff distribution is inspected using the industrial camera. If the inspection meets the standards, the garments proceed to the temperature setting stage; if not, the airflow time is extended, and the inspection is repeated until uniform distribution is achieved.

[0016] The dryer reads the clothing label to obtain material information and sets the temperature and drying time of the low-temperature dryer according to the characteristics of plush clothing. The clothes are placed in the dryer and the drying program is started. The actual temperature and humidity data are compared with the set values. When the temperature deviates from the set value, the power of the dryer heating element is automatically adjusted. When the humidity is not up to standard, the drying time is adjusted. After the drying time is over, the dryer is automatically turned off. At this time, the plush of the clothes has been evenly shaped.

[0017] In conjunction with the first aspect, in the third embodiment of the first aspect of this application, the steps of acquiring clothing images, performing image preprocessing, evaluating the clothing using a deep learning-based fluff density assessment model, and issuing a cleanliness assessment command when it is determined that the uniformity of the clothing fluff meets the standard include:

[0018] The system controls an electric pan-tilt head to drive an industrial camera to capture images of the garment from multiple angles, and performs image preprocessing. A deep learning-based fluff density assessment model, specifically an improved U-Net network, performs pixel-by-pixel analysis on the preprocessed images. The model identifies fluff regions, calculates fluff density at different locations, and compares the fluff density with a preset uniformity standard. If the difference in fluff density between different areas of the garment is within the allowable range, and there is no fluff aggregation or sparseness, the fluff uniformity is deemed to meet the standard; otherwise, the fluff uniformity is deemed not to meet the standard. When the fluff uniformity is determined to meet the standard, a cleanliness assessment command is issued.

[0019] In conjunction with the first aspect, in the fourth embodiment of the first aspect of this application, the step of extracting multispectral features from the clothing image and identifying stains using the YOLOv5 model when a cleanliness assessment instruction is received includes:

[0020] The clothing image was converted to the HSV color space, combined with the original RGB channels, and wavelet transform technology was used to decompose the image into different frequency sub-bands to extract the texture details of the lint and stains, including fiber density and stain edges, forming a multispectral feature dataset.

[0021] The clothing images are combined with a multispectral feature dataset and input into a pre-trained YOLOv5 model. This model is trained based on a plush clothing stain dataset, which contains several stain types. The model scans the clothing images region by region, extracts image features through a convolutional neural network, uses an anchor box mechanism to predict areas where stains may exist, combines a classifier to determine the stain type, and outputs a stain recognition result that includes the stain location and stain type.

[0022] In conjunction with the first aspect, in the fifth embodiment of the first aspect of this application, the step of calculating the stain area ratio based on the stain identification result, classifying the stain severity, and calculating the cleanliness score includes:

[0023] Based on the stain recognition results, the bounding box coordinates of each stain area are determined according to the stain location. The number of pixels within the bounding box is added to obtain the total number of stain pixels. The total number of pixels in the clothing image is calculated. The total number of stain pixels is divided by the total number of pixels in the clothing to obtain the stain area ratio. Each identified stain type is classified according to a preset stain type and severity grading table, and a corresponding weight is assigned to each severity level. For each type of stain, its stain area ratio is multiplied by the corresponding severity weight to obtain the weighted stain area. The weighted stain areas of all categories are added together to obtain the total weighted stain area of ​​the clothing.

[0024] A baseline for calculating the cleanliness score is established. For every increase in the total weighted stain area by a certain percentage, a corresponding score is deducted. Based on the total weighted stain area, the corresponding score is deducted from the baseline proportionally to obtain the cleanliness score.

[0025] In conjunction with the first aspect, in the sixth embodiment of the first aspect of this application, the step of obtaining the physical properties of the fluff, simulating the reflection and occlusion of light on the fluff surface using a ray tracing algorithm, dividing the simulated shadow image into regions, calculating the average color value of each region, and obtaining the natural shadow color of the fluff includes:

[0026] The physical properties of the velvet fibers are obtained from the clothing label, and the lighting conditions in the simulated environment are determined, including the type of light source, light intensity, and light angle. Based on these physical properties, a 3D model of the velvet fibers is constructed using a ray tracing algorithm. The length, curvature, and arrangement of the velvet fibers are input into the 3D model to recreate the realistic velvet structure. The ray tracing algorithm is then activated, emitting light rays from the light source position and allowing the light rays to interact with the constructed 3D model. When the light rays encounter the velvet fibers, the reflection direction of the light rays is calculated based on the reflectivity parameters of the velvet fibers. When the light rays are blocked by the velvet fibers, the area forming a shadow is recorded. The light rays are emitted repeatedly, and through multiple simulation calculations, a shadow image of the velvet fibers under the set lighting conditions is generated.

[0027] The shadow image is divided into multiple regions using a regular grid. For each region, the color values ​​of all pixels in that region are extracted, and the average color value of all pixels in that region is calculated to obtain the average color value of that region. By traversing all the regions, the average color value of each region in the entire shadow image is calculated to obtain the natural shadow color of the fur.

[0028] In conjunction with the first aspect, in the seventh embodiment of the first aspect of this application, the step of converting the stain color and the natural shadow color of the velvet in the stain database to the CIE Lab color space and calculating the color difference includes:

[0029] The stain colors and natural shading colors of the velvet in the stain database are converted to the CIE Lab color space to obtain the brightness, red-green axis, and yellow-blue axis component values ​​corresponding to the stain colors, forming a CIE Lab dataset of stain colors. Based on the location of stains in the clothing images, the stain colors are matched and grouped with the corresponding natural shading colors of the velvet areas. For each group of stain colors and natural shading colors, the CIEDE2000 color difference formula is used for calculation. The brightness, red-green axis, and yellow-blue axis component values ​​of each group of colors are substituted into the calculation process, and the color difference between the colors in the group is obtained through comparison and calculation. This step is repeated to calculate the color difference of all groups.

[0030] In conjunction with the first aspect, in the eighth embodiment of the first aspect of this application, the calculation of the influence value of color difference on the cleanliness score and the adjustment of the cleanliness score include:

[0031] Based on the degree of impact of stains on the cleanliness of clothing, basic weights are assigned to different types of stains; color differences are divided into different levels, and the basic weight of the stain type is multiplied by the color difference level to obtain the comprehensive weight of each stain area; the area ratio of each stain area is multiplied by its corresponding comprehensive weight to obtain the impact value of color difference on the cleanliness score of that area; the impact values ​​of all stain areas are added together to obtain the total impact value of color difference on the overall cleanliness score of clothing.

[0032] The total impact value is converted into a deduction score. The deduction score is then subtracted from the original cleanliness score to obtain the adjusted cleanliness score.

[0033] Secondly, this application provides a cloud-based intelligent assessment system for clothing cleanliness, including:

[0034] The garment lint uniformization module includes an intelligent cleaning unit and a garment lint uniformization unit. The intelligent cleaning unit identifies the garment label, determines that the garment is made of lint, retrieves the corresponding cleaning parameters, and performs intelligent cleaning. The garment lint uniformization unit performs physical combing control, airflow-assisted control, and temperature shaping control on the garment to achieve uniform lint uniformity.

[0035] The instruction issuing module includes a clothing lint uniformity assessment unit and a cleanliness assessment instruction issuing unit. The clothing lint uniformity assessment unit acquires clothing images, performs image preprocessing, and uses a deep learning-based lint density assessment model to assess the clothing. When the cleanliness assessment instruction issuing unit determines that the clothing lint uniformity meets the standard, it issues a cleanliness assessment instruction.

[0036] The cleanliness score calculation module includes a stain recognition unit and a cleanliness score calculation unit. When the stain recognition unit receives a cleanliness assessment instruction, it extracts multispectral features from the clothing image and uses the YOLOv5 model to identify stains. The cleanliness score calculation unit calculates the stain area ratio based on the stain recognition results, classifies the stain severity, and calculates the cleanliness score.

[0037] The cleanliness score adjustment module includes: a natural shadow color calculation unit for lint, a color difference calculation unit, and a cleanliness score adjustment unit. The natural shadow color calculation unit acquires the physical properties of the lint, uses a ray tracing algorithm to simulate the reflection and occlusion of light on the lint surface, divides the simulated shadow image into regions, calculates the average color value of each region, and obtains the natural shadow color of the lint. The color difference calculation unit converts the stain colors and natural shadow colors of the lint from the stain database to the CIE Lab color space and calculates the color difference. The cleanliness score adjustment unit calculates the impact of the color difference on the cleanliness score and adjusts the cleanliness score accordingly.

[0038] Compared with the prior art, the beneficial effects of the present invention are:

[0039] 1. This invention obtains the physical properties of the lint and uses a ray tracing algorithm to simulate natural shadows. Combined with color difference calculation in the CIELab color space, it can accurately distinguish between lint shadows and stains, significantly improving the accuracy of stain identification. At the same time, the lint density evaluation model based on deep learning can monitor lint uniformity in real time, providing a more reliable data foundation for cleanliness assessment.

[0040] 2. This invention provides fully automated control of the entire process, from clothing label recognition, automatic retrieval of washing parameters, and intelligent washing, to lint uniformization and cleanliness assessment. This reduces manual intervention, improves processing efficiency, lowers labor costs, and avoids operational errors caused by human factors.

[0041] 3. This invention effectively achieves uniform distribution of fluff through the coordinated control of physical combing, airflow assistance, and temperature setting, thereby improving the appearance and texture of clothing; precise cleanliness assessment ensures that the degree of cleaning of clothing is just right, avoiding over-washing and damaging the clothing, while thoroughly removing stains. Attached Figure Description

[0042] Figure 1 This is a schematic diagram illustrating the steps of the cloud computing-based intelligent assessment method for clothing cleanliness according to the present invention.

[0043] Figure 2 This is a system structure diagram of the cloud computing-based intelligent assessment system for clothing cleanliness according to the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Example: Figures 1-2 As shown, the present invention provides a technical solution.

[0046] like Figure 1 The schematic diagram illustrates the steps of a cloud-based intelligent assessment method for clothing cleanliness. This application provides a cloud-based intelligent assessment method for clothing cleanliness, including the following steps:

[0047] Step S100: Identify the clothing label. The clothing material is plush. Retrieve the corresponding washing parameters and perform intelligent washing. Perform physical combing control, airflow-assisted control and temperature setting control on the clothing to make the pile of the clothing uniform.

[0048] Specifically, the system identifies clothing tags using RFID readers or QR code scanners; extracts the material field to determine if the clothing is made of plush or fleece; if it is, it proceeds to the next step; otherwise, it retrieves the corresponding material's cleaning parameters and executes other cleaning processes; it retrieves the cleaning parameters for plush clothing from a preset parameter database and sends start commands to the ultrasonic cleaner and detergent dispenser; during the cleaning process, it monitors the operating status of the cleaning equipment in real time; after the cleaning time is completed, it triggers the rinsing program; it uses a conductivity sensor to detect the conductivity of the cleaning solution to determine the amount of detergent residue; when the conductivity reaches the preset rinsing end standard, the water quality is considered to be up to standard, the rinsing program is stopped, and the intelligent cleaning process is completed.

[0049] Furthermore, an industrial camera is used to scan the garment, and image recognition technology is used to locate areas where fibers are tightly packed, marking the areas that need to be combed. An electric soft brush device is then moved above the garment, and a servo motor adjusts the brush head angle according to the marked area. A pressure sensor at the end of the brush provides real-time feedback on the contact pressure with the garment, adjusting the servo motor torque to maintain a stable combing force. The brush is controlled to move at a constant speed, starting from one end of the garment and inserting into the fiber gaps for combing. After combing a fixed distance, an ion air gun is activated to blow away dust and impurities carried out during the combing process, preventing them from re-adhering to the fibers, and the combing continues to the next section. The above steps are repeated until all marked areas have been combed.

[0050] After combing, the garments are hung in designated locations. The motorized gimbal automatically adjusts the blower outlet based on the seam angle and turns on the blower. An industrial camera captures real-time images of the fluff's movement under the airflow, analyzing the uniformity of fluff distribution. If uneven fluff distribution is detected, the blower speed is increased; if fluff scatters, the speed is decreased. After the blower continues blowing for a period, it is paused, and the fluff distribution is inspected using the industrial camera. If the inspection meets the standards, the garments proceed to the temperature setting stage; if not, the airflow time is extended, and the inspection is repeated until uniform distribution is achieved.

[0051] The dryer reads the clothing label to obtain material information and sets the temperature and drying time of the low-temperature dryer according to the characteristics of plush clothing. The clothes are placed in the dryer and the drying program is started. The actual temperature and humidity data are compared with the set values. When the temperature deviates from the set value, the power of the dryer heating element is automatically adjusted. When the humidity is not up to standard, the drying time is adjusted. After the drying time is over, the dryer is automatically turned off. At this time, the plush of the clothes has been evenly shaped.

[0052] Step S200: Acquire images of clothing, perform image preprocessing, evaluate clothing using a deep learning-based fluff density evaluation model, and issue a cleanliness evaluation command when it is determined that the uniformity of the clothing fluff meets the standard.

[0053] Specifically, the system controls an electric pan-tilt head to drive an industrial camera to capture images of the clothing from multiple angles, and performs image preprocessing. A deep learning-based lint density assessment model, specifically an improved U-Net network, performs pixel-by-pixel analysis on the preprocessed images. The model identifies lint regions, calculates lint density at different locations, and compares the lint density with a preset uniformity standard. When the lint density difference between different areas of the clothing is within the allowable range, and there is no lint aggregation or sparseness, the lint uniformity is deemed to meet the standard; otherwise, the lint uniformity is deemed not to meet the standard. When the lint uniformity is determined to meet the standard, a cleanliness assessment command is issued.

[0054] In one specific embodiment, a woolen plush garment is selected (the plush at the collar and seams tends to gather), and the plush uniformity standard is set as follows: the difference in plush density in each area is ≤10%, and there are no obvious plush accumulation or sparse areas.

[0055] The equipment parameters are as follows: industrial camera with a resolution of 4K (3840×2160 pixels), equipped with a 100mm macro lens, a ring-shaped shadowless light source with a brightness of 4000 lux, and a shooting distance of 30cm.

[0056] Shooting angles were as follows: vertical overhead shot: to capture the overall distribution of the pile on the garment; left 45° side shot: to focus on the fiber stacking area at the seam; right 45° side shot: to supplement the shooting of the hidden areas of the folds; and top 30° upward shot: to check the deep pile condition of the neckline.

[0057] Three images were captured from each angle, for a total of 12 images, with each image being approximately 20MB in size.

[0058] The nonlocal mean denoising algorithm was used to eliminate reflective noise on the velvet surface, improving the signal-to-noise ratio of the processed image by 35%. The MSR algorithm was used to improve local contrast, increasing the color difference between velvet and stains from ΔE=3.2 to ΔE=5.8, and improving the clarity of fiber gaps at the stitching by 40%. The images were uniformly scaled to 640×640 pixels for easier model analysis.

[0059] The improved U-Net network was used to evaluate the fluff density. The model input was a preprocessed 640×640 pixel image with three RGB channels. The model output was a pixel-by-pixel segmented fluff region mask (white for fluff, black for background) and fluff density values ​​for each region (unit: roots / square millimeter). The encoder extracted fluff texture features (such as fiber curvature and gaps), and the decoder reconstructed the pixel-level segmentation results. A CBAM attention mechanism was introduced to enhance feature extraction in dense areas such as creases, resulting in an 18% improvement in recognition accuracy compared to the original U-Net.

[0060] The image was divided into 10×10=100 grid regions, each region being 64×64 pixels. In the neckline region, the average pile density was 28.5 fibers / mm², with a maximum density of 31.2 fibers / mm² and a minimum density of 25.8 fibers / mm². The density difference was approximately 19% (31.2-25.8) / 28.5 ≈ 19% (unprocessed initially). In the quilting region, the initial density difference reached 25%, with localized high-density areas (35.6 fibers / mm²) and sparse areas (22.1 fibers / mm²) caused by fiber stacking. After physical combing, airflow assistance, and temperature setting, the density difference in the neckline region decreased to 8.7% (29.1-26.5 fibers / mm²), and the density difference in the quilting region decreased to 9.3% (28.8-25.7 fibers / mm²), both ≤10%. The model did not detect any fiber aggregation (such as clumps of fibers) or sparseness (visible base fabric).

[0061] When the difference in fluff density in all detection areas is ≤10% and there is no abnormal distribution, the system determines that "fluff uniformity meets the standard" and sends an instruction to the cleanliness assessment module. In the experiment, the wool garment met the standard after the third uniformization treatment, which took about 12 minutes (including 3 image acquisitions and assessments).

[0062] Step S300: When a cleanliness assessment instruction is received, multispectral features are extracted from the clothing image, and stains are identified using the YOLOv5 model; based on the stain identification results, the stain area ratio is calculated, the stain severity is graded, and a cleanliness score is calculated.

[0063] Specifically, the clothing image is converted to the HSV color space, combined with the original RGB channels, and wavelet transform technology is used to decompose the image into different frequency sub-bands to extract the texture details of the lint and stains, including fiber density and stain edges, forming a multispectral feature dataset.

[0064] The clothing images are combined with a multispectral feature dataset and input into a pre-trained YOLOv5 model. This model is trained based on a plush clothing stain dataset, which contains several stain types. The model scans the clothing images region by region, extracts image features through a convolutional neural network, uses an anchor box mechanism to predict areas where stains may exist, combines a classifier to determine the stain type, and outputs a stain recognition result that includes the stain location and stain type.

[0065] Furthermore, based on the stain recognition results, the bounding box coordinates of each stain area are determined according to the stain location. The number of pixels within the bounding box is added to obtain the total number of stain pixels. The total number of pixels in the clothing image is calculated. The total number of stain pixels is divided by the total number of pixels in the clothing to obtain the stain area ratio. Each identified stain type is classified according to a preset stain type and severity grading table, and a corresponding weight is assigned to each severity level. For each type of stain, its stain area ratio is multiplied by the corresponding severity weight to obtain the weighted stain area. The weighted stain areas of all categories are added together to obtain the total weighted stain area of ​​the clothing.

[0066] A baseline for calculating the cleanliness score is established. For every increase in the total weighted stain area by a certain percentage, a corresponding score is deducted. Based on the total weighted stain area, the corresponding score is deducted from the baseline proportionally to obtain the cleanliness score.

[0067] In one specific embodiment, a woolen coat (stained with juice, oil and dust) is selected. The preset cleanliness score is 100 points. For every 1% increase in the total weighted stain area, 1 point is deducted. The severity of stains is graded as follows: dust (light, weight 1), juice stain (moderate, weight 2), and oil stain (severe, weight 3).

[0068] The original 4K RGB image (3840×2160 pixels) was converted to the HSV color space, and the hue (H), saturation (S), and lightness (V) channels were obtained. Oil stains exhibited high saturation characteristics in the S channel (S value > 80), while juice stains were concentrated in the 35°-45° range in the H channel. The image was decomposed into low-frequency (approximate components) and high-frequency (detail components) sub-bands using the db4 wavelet, and the velvet texture (fiber density in the low-frequency region) and stain edges (gradient changes in the high-frequency region) were extracted. The features from the RGB three channels, HSV three channels, and the four sub-bands after wavelet decomposition were integrated to form a 10-channel multispectral feature matrix with dimensions of 640×640×10.

[0069] Using a dataset containing 100,000 images of plush clothing, the stains are labeled with three types: dust, juice stains, and oil stains. The anchor frame size is optimized for the gaps between plush fibers (e.g., the minimum anchor frame is 16×16 pixels to accommodate tiny stains at the seams).

[0070] The input image is: after preprocessing, the size is 640×640 pixels, and the multispectral features are normalized to [-1,1].

[0071] The test results are as follows:

[0072] One oil stain was identified at the collar, with bounding box coordinates of (150, 80) - (280, 160) and pixel count of 130 × 80 = 10400.

[0073] Two juice stains were identified at the cuff, with bounding box pixel counts of 8000 and 5000 respectively;

[0074] Dust and dirt can be detected at the hem seam, with a resolution of 3000 pixels.

[0075] Calculate the stain area percentage: Total number of pixels on clothing: 640×640=409600; Total number of pixels on stain: 10400+8000+5000+3000=26400; Area percentage: 26400 / 409600≈6.45%.

[0076] The severity of stains is graded as follows: oil stains: heavy stains, weight 3; juice stains: moderate stains, weight 2; dust: light stains, weight 1.

[0077] The weighted stain areas are as follows: oil stains: (10400 / 409600)×3≈0.075×3=0.225; juice stains: (8000+5000) / 409600×2≈0.0318×2=0.0636; dust: 3000 / 409600×1≈0.0073×1=0.0073; total weighted stain area: 0.225+0.0636+0.0073≈0.296 (i.e. 29.6%).

[0078] The deduction rule is as follows: 1 point is deducted for every 1% of the total weighted stain area, and 0.296% is calculated as 0.3%; the score is calculated as follows: 100 - 0.3 × 1 = 99.7 points (rounded to the nearest whole number).

[0079] Step S400: Obtain the physical properties of the velvet, use a ray tracing algorithm to simulate the reflection and occlusion of light on the velvet surface, divide the simulated shadow image into regions, calculate the average color value of each region, and obtain the natural shadow color of the velvet; convert the stain color and the natural shadow color of the velvet in the stain database to the CIE Lab color space, and calculate the color difference; calculate the impact of the color difference on the cleanliness score, and adjust the cleanliness score accordingly.

[0080] Specifically, the physical properties of the velvet fibers are obtained from the clothing label, and the lighting conditions in the simulated environment are determined, including the type of light source, light intensity, and light angle. Based on these physical properties, a three-dimensional model of the velvet fibers is constructed using a ray tracing algorithm. The length, curvature, and arrangement of the velvet fibers are input into the three-dimensional model to recreate the realistic velvet structure. The ray tracing algorithm is then activated, emitting light rays from the light source position and allowing the light rays to interact with the constructed three-dimensional model. When the light rays encounter the velvet fibers, the reflection direction of the light rays is calculated based on the reflectivity parameters of the velvet fibers. When the light rays are blocked by the velvet fibers, the area forming a shadow is recorded. The light rays are emitted repeatedly, and through multiple simulation calculations, a shadow image of the velvet fibers under the set lighting conditions is generated.

[0081] The shadow image is divided into multiple regions using a regular grid. For each region, the color values ​​of all pixels in that region are extracted, and the average color value of all pixels in that region is calculated to obtain the average color value of that region. By traversing all the regions, the average color value of each region in the entire shadow image is calculated to obtain the natural shadow color of the fur.

[0082] Furthermore, the stain colors and natural shading colors of the velvet in the stain database are converted to the CIE Lab color space to obtain the brightness, red-green axis, and yellow-blue axis component values ​​corresponding to the stain colors, forming a CIE Lab dataset of stain colors. Based on the location of the stains in the clothing images, the stain colors are matched and grouped with the corresponding natural shading colors of the velvet areas. For each group of stain colors and natural shading colors, the CIEDE2000 color difference formula is used for calculation. The brightness, red-green axis, and yellow-blue axis component values ​​of each group of colors are substituted into the calculation process. Through comparison and calculation, the color difference between the colors in that group is obtained. This step is repeated to calculate the color difference of all groups.

[0083] Furthermore, based on the degree of impact of stains on the cleanliness of clothing, basic weights are assigned to different types of stains; color differences are divided into different levels, and the basic weight of the stain type is multiplied by the color difference level to obtain the comprehensive weight of each stain area; the area ratio of each stain area is multiplied by its corresponding comprehensive weight to obtain the influence value of color difference on the cleanliness score of that area; the influence values ​​of all stain areas are added together to obtain the total influence value of color difference on the overall cleanliness score of clothing.

[0084] The total impact value is converted into a deduction score. The deduction score is then subtracted from the original cleanliness score to obtain the adjusted cleanliness score.

[0085] In one specific embodiment, a cashmere scarf was selected, with coffee stains (moderate contamination) and a small amount of machine oil residue (heavy contamination) on its surface, and there was interference from the fuzzy shadow at the stitching. The preset cleanliness score benchmark was 100 points, and the original cleanliness score was 92 points (calculated based on step S300). The basic weights for stain types were: coffee stains (2), machine oil (3); the weights for color difference grading were: slight (0.3), moderate (0.6), and obvious (0.9).

[0086] The fluff parameters were obtained from the clothing label: average length 12mm, diameter 18μm, average bending angle 45°, and arrangement density 2000 fibers / cm². Spectral reflectance parameters: approximately 75%-85% in the visible light band (400-700nm). A parallel light source was set with an intensity of 1000 lux and an illumination angle of 45° (simulating natural oblique light). In the ray tracing algorithm, a highly realistic 3D model of the fluff was constructed based on its length, bending shape, and arrangement. Individual fibers were simulated as cylinders and randomly bent, forming a fluffy structure overall. 100,000 rays were emitted to simulate illumination. When a ray touched the fluff, the reflection direction was calculated based on the reflectance (e.g., at 80% reflectance, 80% of the light is reflected and 20% is absorbed); obscured areas were recorded as shadow areas. After multiple iterations of calculation, a velvet shadow image was generated with a resolution of 640×640 pixels. The shadow area is dark gray (RGB:80,80,80), which contrasts with the light gray of the velvet body (RGB:120,120,120).

[0087] Using a 20×20 regular grid, the image was divided into 400 regions, each region being 32×32 pixels. Region 1: RGB values ​​of 32×32=1024 pixels were extracted, and the average value was calculated to be (82,82,82); Region 2: The average value was (85,85,85); After traversing all regions, the average RGB value of the natural shadow color of the fur was determined to be (83,83,83), which was converted to the CIE Lab color space as (L*:32,a*:-1,b*:1).

[0088] The original RGB(150,75,0) of the coffee stain was converted to CIE Lab(L*:38,a*:25,b*:50); the RGB(50,50,50) of the engine oil stain was converted to CIE Lab(L*:20,a*:0,b*:0). The coffee stain was grouped with the corresponding area's fuzzy shadow color, and the engine oil stain was grouped similarly. The color difference between the coffee stain and the shadow (CIEDE2000): ΔE≈12 (significant color difference, weight 0.9); the color difference between the engine oil stain and the shadow: ΔE≈18 (significant color difference, weight 0.9).

[0089] Perform comprehensive weight calculation:

[0090] Coffee stains: Base weight 2 × Color difference weight 0.9 = 1.8;

[0091] Engine oil stains: base weight 3 × color difference weight 0.9 = 2.7.

[0092] Perform the impact value calculation:

[0093] The coffee stain area accounts for 1.2% of the total area, and its impact value is 1.2% × 1.8 = 0.0216.

[0094] The area affected by the oil stain is 0.8%, and its impact value is 0.8% × 2.7 = 0.0216.

[0095] Total impact value = 0.0216 + 0.0216 = 0.0432 (i.e. 4.32%).

[0096] Deduction of 4.32 points, adjusted cleanliness score = 92 - 4.32 = 87.68 points.

[0097] like Figure 2 The system architecture diagram of the cloud-based intelligent clothing cleanliness assessment system is shown in this application. The system includes:

[0098] The garment lint uniformization module includes an intelligent cleaning unit and a garment lint uniformization unit. The intelligent cleaning unit identifies the garment label, determines that the garment is made of lint, retrieves the corresponding cleaning parameters, and performs intelligent cleaning. The garment lint uniformization unit performs physical combing control, airflow-assisted control, and temperature shaping control on the garment to achieve uniform lint uniformity.

[0099] The instruction issuing module includes a clothing lint uniformity assessment unit and a cleanliness assessment instruction issuing unit. The clothing lint uniformity assessment unit acquires clothing images, performs image preprocessing, and uses a deep learning-based lint density assessment model to assess the clothing. When the cleanliness assessment instruction issuing unit determines that the clothing lint uniformity meets the standard, it issues a cleanliness assessment instruction.

[0100] The cleanliness score calculation module includes a stain recognition unit and a cleanliness score calculation unit. When the stain recognition unit receives a cleanliness assessment instruction, it extracts multispectral features from the clothing image and uses the YOLOv5 model to identify stains. The cleanliness score calculation unit calculates the stain area ratio based on the stain recognition results, classifies the stain severity, and calculates the cleanliness score.

[0101] The cleanliness score adjustment module includes: a natural shadow color calculation unit for lint, a color difference calculation unit, and a cleanliness score adjustment unit. The natural shadow color calculation unit acquires the physical properties of the lint, uses a ray tracing algorithm to simulate the reflection and occlusion of light on the lint surface, divides the simulated shadow image into regions, calculates the average color value of each region, and obtains the natural shadow color of the lint. The color difference calculation unit converts the stain colors and natural shadow colors of the lint from the stain database to the CIE Lab color space and calculates the color difference. The cleanliness score adjustment unit calculates the impact of the color difference on the cleanliness score and adjusts the cleanliness score accordingly.

[0102] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A cloud computing-based intelligent assessment method for clothing cleanliness, characterized in that, Includes the following steps: The system identifies the clothing label, determines that the clothing is made of plush material, retrieves the corresponding washing parameters, and performs intelligent washing. It also performs physical combing control, airflow-assisted control, and temperature setting control on the clothing to ensure that the pile of the clothing is evenly distributed. The system collects images of clothing, performs image preprocessing, and uses a deep learning-based fluff density assessment model to evaluate the clothing. When it is determined that the uniformity of the fluff in the clothing meets the standard, a cleanliness assessment command is issued. When a cleanliness assessment instruction is received, multispectral features are extracted from the clothing image, and stains are identified using the YOLOv5 model. Based on the stain identification results, the stain area ratio is calculated, the stain severity is graded, and a cleanliness score is calculated. The physical properties of the velvet fibers are obtained, and ray tracing algorithms are used to simulate the reflection and occlusion of light on the velvet surface. The simulated shadow image is then divided into regions, and the average color value of each region is calculated to obtain the natural shadow color of the velvet fibers, including: The physical properties of the velvet fibers are obtained from the clothing label, and the lighting conditions in the simulated environment are determined, including the type of light source, light intensity, and light angle. Based on these physical properties, a 3D model of the velvet fibers is constructed using a ray tracing algorithm. The length, curvature, and arrangement of the velvet fibers are input into the 3D model to recreate the realistic velvet structure. The ray tracing algorithm is then activated, emitting light rays from the light source position and allowing the light rays to interact with the constructed 3D model. When the light rays encounter the velvet fibers, the reflection direction of the light rays is calculated based on the reflectivity parameters of the velvet fibers. When the light rays are blocked by the velvet fibers, the area forming a shadow is recorded. The light rays are emitted repeatedly, and through multiple simulation calculations, a shadow image of the velvet fibers under the set lighting conditions is generated. The shadow image is divided into multiple regions using a regular grid. For each region, the color values ​​of all pixels within that region are extracted, and the average color value of all pixels within that region is calculated to obtain the average color value of that region. By traversing all the regions, the average color value of each region in the entire shadow image is calculated to obtain the natural shadow color of the fur. Convert stain colors and natural velvet shadow colors from the stain database to the CIE Lab color space, and calculate color differences, including: The stain colors and natural shading colors of the velvet in the stain database are converted to the CIE Lab color space to obtain the brightness, red-green axis, and yellow-blue axis component values ​​corresponding to the stain colors, forming a CIE Lab dataset of stain colors. Based on the location of stains in the clothing images, the stain colors are matched and grouped with the corresponding natural shading colors of the velvet areas. For each group of stain colors and natural shading colors, the CIEDE2000 color difference formula is used for calculation. The brightness, red-green axis, and yellow-blue axis component values ​​of each color are substituted into the calculation process, and the color difference between the colors in that group is obtained through comparison and calculation. This step is repeated to calculate the color difference for all groups. Calculate the impact of color difference on the cleanliness score, and adjust the cleanliness score accordingly, including: Based on the degree of impact of stains on the cleanliness of clothing, basic weights are assigned to different types of stains; color differences are divided into different levels, and the basic weight of the stain type is multiplied by the color difference level to obtain the comprehensive weight of each stain area; the area ratio of each stain area is multiplied by its corresponding comprehensive weight to obtain the impact value of color difference on the cleanliness score of that area; the impact values ​​of all stain areas are added together to obtain the total impact value of color difference on the overall cleanliness score of clothing. The total impact value is converted into a deduction score. The deduction score is then subtracted from the original cleanliness score to obtain the adjusted cleanliness score.

2. The intelligent assessment method for clothing cleanliness based on cloud computing according to claim 1, characterized in that, The process involves identifying clothing tags, determining if the clothing is made of fleece or similar materials, retrieving corresponding washing parameters, and performing intelligent washing, including: The system identifies clothing tags using RFID readers or QR code scanners; extracts the material field to determine if the clothing is made of plush or fleece; if it is, it proceeds to the next step; otherwise, it retrieves the corresponding material cleaning parameters and executes other cleaning processes; it retrieves the cleaning parameters for plush clothing from a preset parameter database and sends start commands to the ultrasonic cleaner and detergent dispenser; during the cleaning process, it monitors the operating status of the cleaning equipment in real time; after the cleaning time is completed, it triggers the rinsing program; it detects the conductivity of the cleaning solution using a conductivity sensor to determine the amount of detergent residue; when the conductivity reaches the preset rinsing end standard, the water quality is considered to be up to standard, the rinsing program is stopped, and the intelligent cleaning process is completed.

3. The intelligent assessment method for clothing cleanliness based on cloud computing according to claim 1, characterized in that, The physical combing control, airflow-assisted control, and temperature-setting control of the garments, which uniformly distribute the fabric's nap, include: An industrial camera scans the garment, and image recognition technology locates areas with tightly packed fibers, marking the areas that need combing. An electric soft brush is then moved over the garment, and a servo motor adjusts the brush head angle according to the marked areas. A pressure sensor at the brush tip provides real-time feedback on the contact pressure with the garment, adjusting the servo motor torque to maintain a stable combing force. The brush moves at a constant speed, starting from one end of the garment and inserting into the fiber gaps for combing. After combing a fixed distance, an ion air gun is activated to blow away dust and impurities carried out during the combing process, preventing them from re-adhering to the fibers, before continuing to the next section. This process is repeated until all marked areas have been combed. After combing, the garments are hung in designated locations. The motorized gimbal automatically adjusts the blower outlet based on the seam angle and turns on the blower. An industrial camera captures real-time images of the fluff's movement under the airflow, analyzing the uniformity of fluff distribution. If uneven fluff distribution is detected, the blower speed is increased; if fluff scatters, the speed is decreased. After the blower continues blowing for a period, it is paused, and the fluff distribution is inspected using the industrial camera. If the inspection meets the standards, the garments proceed to the temperature setting stage; if not, the airflow time is extended, and the inspection is repeated until uniform distribution is achieved. The dryer reads the clothing label to obtain material information and sets the temperature and drying time of the low-temperature dryer according to the characteristics of plush clothing. The clothes are placed in the dryer and the drying program is started. The actual temperature and humidity data are compared with the set values. When the temperature deviates from the set value, the power of the dryer heating element is automatically adjusted. When the humidity is not up to standard, the drying time is adjusted. After the drying time is over, the dryer is automatically turned off. At this time, the plush of the clothes has been evenly shaped.

4. The intelligent assessment method for clothing cleanliness based on cloud computing according to claim 1, characterized in that, The process involves acquiring images of the clothing, performing image preprocessing, and using a deep learning-based fluff density assessment model to evaluate the clothing. When the uniformity of the clothing fluff is determined to meet the standard, a cleanliness assessment command is issued, including: The system controls an electric pan-tilt head to drive an industrial camera to capture images of the garment from multiple angles, and performs image preprocessing. A deep learning-based fluff density assessment model, specifically an improved U-Net network, performs pixel-by-pixel analysis on the preprocessed images. The model identifies fluff regions, calculates fluff density at different locations, and compares the fluff density with a preset uniformity standard. If the difference in fluff density between different areas of the garment is within the allowable range, and there is no fluff aggregation or sparseness, the fluff uniformity is deemed to meet the standard; otherwise, the fluff uniformity is deemed not to meet the standard. When the fluff uniformity is determined to meet the standard, a cleanliness assessment command is issued.

5. The intelligent assessment method for clothing cleanliness based on cloud computing according to claim 1, characterized in that, When a cleanliness assessment instruction is received, multispectral feature extraction is performed on the clothing image, and stain identification is performed using the YOLOv5 model, including: The clothing image was converted to the HSV color space, combined with the original RGB channels, and wavelet transform technology was used to decompose the image into different frequency sub-bands to extract the texture details of the lint and stains, including fiber density and stain edges, forming a multispectral feature dataset. The clothing images are combined with a multispectral feature dataset and input into a pre-trained YOLOv5 model. This model is trained based on a plush clothing stain dataset, which contains several stain types. The model scans the clothing images region by region, extracts image features through a convolutional neural network, uses an anchor box mechanism to predict areas where stains may exist, combines a classifier to determine the stain type, and outputs a stain recognition result that includes the stain location and stain type.

6. The intelligent assessment method for clothing cleanliness based on cloud computing according to claim 1, characterized in that, The process of calculating the stain area percentage based on stain identification results, classifying stain severity, and calculating a cleanliness score includes: Based on the stain recognition results, the bounding box coordinates of each stain area are determined according to the stain location. The number of pixels within the bounding box is added to obtain the total number of stain pixels. The total number of pixels in the clothing image is calculated. The total number of stain pixels is divided by the total number of pixels in the clothing to obtain the stain area ratio. Each identified stain type is classified according to a preset stain type and severity grading table, and a corresponding weight is assigned to each severity level. For each type of stain, its stain area ratio is multiplied by the corresponding severity weight to obtain the weighted stain area. The weighted stain areas of all categories are added together to obtain the total weighted stain area of ​​the clothing. A baseline for calculating the cleanliness score is established. For every increase in the total weighted stain area by a certain percentage, a corresponding score is deducted. Based on the total weighted stain area, the corresponding score is deducted from the baseline proportionally to obtain the cleanliness score.

7. A cloud-based intelligent assessment system for clothing cleanliness, using the cloud-based intelligent assessment method for clothing cleanliness as described in any one of claims 1-6, characterized in that, include: The garment lint uniformization module includes an intelligent cleaning unit and a garment lint uniformization unit. The intelligent cleaning unit identifies the garment label, determines that the garment is made of lint, retrieves the corresponding cleaning parameters, and performs intelligent cleaning. The garment lint uniformization unit performs physical combing control, airflow-assisted control, and temperature shaping control on the garment to achieve uniform lint uniformity. The instruction issuing module includes a clothing lint uniformity assessment unit and a cleanliness assessment instruction issuing unit. The clothing lint uniformity assessment unit acquires clothing images, performs image preprocessing, and uses a deep learning-based lint density assessment model to assess the clothing. When the cleanliness assessment instruction issuing unit determines that the clothing lint uniformity meets the standard, it issues a cleanliness assessment instruction. The cleanliness score calculation module includes a stain recognition unit and a cleanliness score calculation unit. When the stain recognition unit receives a cleanliness assessment instruction, it extracts multispectral features from the clothing image and uses the YOLOv5 model to identify stains. The cleanliness score calculation unit calculates the stain area ratio based on the stain recognition results, classifies the stain severity, and calculates the cleanliness score. The cleanliness score adjustment module includes: a natural shadow color calculation unit for lint, a color difference calculation unit, and a cleanliness score adjustment unit. The natural shadow color calculation unit acquires the physical properties of the lint, uses a ray tracing algorithm to simulate the reflection and occlusion of light on the lint surface, divides the simulated shadow image into regions, calculates the average color value of each region, and obtains the natural shadow color of the lint. The color difference calculation unit converts the stain colors and natural shadow colors of the lint from the stain database to the CIE Lab color space and calculates the color difference. The cleanliness score adjustment unit calculates the impact of the color difference on the cleanliness score and adjusts the cleanliness score accordingly.

Images