An evaluation method for the adhesion performance of aggregate and asphalt

Abstract

An aggregate asphalt adhesion rotary tester and its adhesion performance evaluation method. At present, during the process of manually dividing the thresholds of the three channels in the RGB color space for different aggregates, the accuracy is poor and the operation process is cumbersome; in the present invention, the test component is arranged in the inner cavity, the first support plate and the second support plate are arranged on the bottom plate, the transmission sprocket assembly is arranged on the first support plate, a plurality of rollers are arranged between the first support plate and the second support plate, one end of each roller is connected to the transmission sprocket assembly, the other end of each roller is hinged to the second support plate, and each roller makes a self-rotation movement between the first support plate and the second support plate under the drive of the transmission sprocket assembly, and a clamping gap for fitting the test bottle of the specimen is formed between every two adjacent rollers; the aggregate asphalt adhesion performance evaluation method is to use the image data collected by the aggregate asphalt adhesion rotary tester as the basic data and conduct a quantitative evaluation process of the aggregate asphalt adhesion performance on the basic data.

Description

Technical Field

[0001] The present invention specifically relates to a method for evaluating the adhesion performance of aggregate to asphalt. Background Art

[0002] Currently, the common methods for evaluating the adhesion of aggregate to asphalt in China include those proposed in the "Test Procedures for Bitumen and Bituminous Mixtures for Highway Engineering" (JTG E20-2011): the boiling water method applicable to aggregates with a maximum particle size greater than 13.2 mm; the water immersion method applicable to aggregates with a maximum particle size less than or equal to 13.2 mm. In contrast, the rotating bottle method has the advantages of dynamically simulating actual working conditions, good repeatability, controllable test conditions, wide application range, simplicity and speed. According to the EU specification EN 12697-11A, the rotating bottle method is generally used in EU countries to evaluate the adhesion of aggregate to asphalt. According to the "Test Procedures for Bitumen and Bituminous Mixtures for Highway Engineering (Draft for Comment)", it is recommended to use the rotating bottle method to evaluate the adhesion of asphalt-aggregate (T 0675-201X). It can be seen that the industry has a relatively high recognition of the rotating bottle method.

[0003] Currently, the evaluation standard of the traditional rotating bottle method for asphalt adhesion is that two experimenters use the visual inspection method to evaluate the coverage rate of asphalt, and its accuracy is 5%. This method has the problems of large error and insufficient accuracy. Currently, the main improvement in the academic circle is to take pictures of the results and manually divide the thresholds of the three channels in the RGB color space for different aggregates. The specific problems are mainly as follows:

[0004] First, the accuracy is poor: the colors of the same type of aggregate are different, and the distinguishability of the background, aggregate, and asphalt is not good. The specific concentrated interference factors are shadows and water. The shadow and asphalt are both black, which will reduce the accuracy during recognition. When the aggregate is taken out of the water, it will inevitably be contaminated with water, and the water will reflect light and affect the results. In addition, the colors of some aggregates are similar to those of asphalt, which will also cause an impact;

[0005] Second, for each type of aggregate, it is necessary to set the threshold in advance, and the process is relatively cumbersome, and redundant steps are difficult to avoid.

[0006] In short, the accuracy of evaluating the coverage rate of asphalt by the visual inspection method is poor, and there is a lack of a standardized and unified evaluation method for the coverage rate of asphalt. Summary of the Invention

[0007] In order to overcome the defects of the prior art, the present invention provides a method for evaluating the adhesion performance of aggregate to asphalt to solve the above problems.

[0008] A rotating tester for the adhesion of aggregate to asphalt, comprising a bottom plate, a housing and a test component. The housing is arranged on the bottom plate, and an inner cavity is formed between the housing and the bottom plate. The test component is arranged in the inner cavity;

[0009] The test component includes a first support plate, a second support plate, a plurality of rollers, a transmission sprocket assembly, and a driving member. The first support plate and the second support plate are vertically arranged in parallel on the bottom plate. The transmission sprocket assembly is arranged on the first support plate and is connected to the driving member. The plurality of rollers are arranged between the first support plate and the second support plate. One end of each roller passes through the first support plate and is connected to the transmission sprocket assembly, and the other end of each roller is hinged to the second support plate. Each roller makes a self-rotation movement between the first support plate and the second support plate driven by the transmission sprocket assembly, and a clamping gap for fitting the test bottle of the specimen is formed between every two adjacent rollers.

[0010] As a preferred solution: A first opening is machined on the outer shell, and an opening and closing cover plate is hinged on the first opening.

[0011] As a preferred solution: A second opening is machined on the outer shell, and a viewing window is arranged on the second opening.

[0012] As a preferred solution: The transmission sprocket assembly includes a chain, an intermediate gear, a driving gear, and a plurality of driven gears. The driven gears are arranged in one-to-one correspondence with the rollers. Each driven gear is sleeved on its corresponding roller. One end of the roller passes through the first support plate and is connected to its corresponding driven gear. An intermediate gear is arranged on the outer wall of the first support plate, and a driving gear is sleeved on the power output shaft of the driving member. A chain is arranged between the driving gear, the intermediate gear, and the plurality of driven gears.

[0013] As a preferred solution: A strip-shaped support plate is arranged on the outer wall of the first support plate along its length direction, and a chain is arranged on the strip-shaped support plate.

[0014] As a preferred solution: It further includes a multi-position irradiation shadowless lamp assembly. The multi-position irradiation shadowless lamp assembly is arranged on the inner wall of the outer shell. The multi-position irradiation shadowless lamp assembly is cooperatively provided with a paving plate. The paving plate is arranged directly below the multi-position irradiation shadowless lamp assembly. The paving plate is arranged on the bottom plate. The multi-position irradiation shadowless lamp assembly includes a main support plate, a plurality of strip-shaped tracks, and a plurality of self-moving light sources. A plurality of strip-shaped tracks are arranged on the main support plate, and a plurality of self-moving light sources are arranged on each strip-shaped track. Each self-moving light source includes a support frame, a motor, a gear, and a light source. The support frame is arranged on its corresponding strip-shaped track. One side of the support frame is slidably matched with one side of the strip-shaped track. A motor is arranged on the other side of the support frame. A gear is sleeved on the output shaft of the motor. The gear meshes with the strip-shaped track. A light source is hinged to the bottom of the support frame. The light-emitting direction of the light source faces downward. The light source reciprocally moves along the length direction of the strip-shaped track driven by the support frame.

[0015] As a preferred solution: It further includes a multi - irradiation shadowless lamp assembly, which is arranged outside the housing. The multi - irradiation shadowless lamp assembly includes a main support plate, a plurality of strip - shaped rails, and a plurality of self - moving light sources. A plurality of strip - shaped rails are arranged at the bottom of the main support plate, and a number of self - moving light sources are arranged on each strip - shaped rail; Each self - moving light source includes a support frame, a motor, a gear, and a light source. The support frame is arranged on its corresponding strip - shaped rail, one side of the support frame is slidably matched with one side of the strip - shaped rail, a motor is arranged on the other side of the support frame, a gear is sleeved on the output shaft of the motor, the gear meshes with the strip - shaped rail, the bottom of the support frame is hinged with a light source, and the light emission direction of the light source is downward. The light source reciprocally moves along the length direction of the strip - shaped rail under the drive of the support frame.

[0016] As a preferred solution: A polyester reflective film sheet is arranged at the bottom of the main support plate, a plurality of strip - shaped rails are arranged in an array form on the polyester reflective film sheet, and an embedded camera is arranged at the center of the bottom of the main support plate.

[0017] A method for evaluating the adhesion performance of aggregate to asphalt is realized by using the above - mentioned aggregate - asphalt adhesion rotary tester. The method for evaluating the adhesion performance of aggregate to asphalt is based on the image data collected by the aggregate - asphalt adhesion rotary tester and conducts a quantitative evaluation process of the adhesion performance of aggregate to asphalt. Specifically:

[0018] The process of quantitative evaluation through the basic data is to form a training set from the image data obtained by the aggregate - asphalt adhesion rotary tester. After randomly cropping a predetermined number of images from the training set to form classifications, the class center point of each class is defined as the arithmetic mean of the pixel coordinates of all pixels in that class. The calculation formula of the arithmetic mean is:

[0019]

[0020] In the above formula, (x i , y i ) is the coordinate of the i - th pixel point in class C, and N C is the number of pixels in class C; Then a weight coefficient w is introduced for each class, and the sampling probability of each class is adjusted according to the corresponding weight coefficient w. The sampling probability P of each class is calculated according to its corresponding weight coefficient w. The calculation formula of the sampling probability P is:

[0021]

[0022] According to the formula of the sampling area of each class, the size of the sampling area is calculated through the weight coefficient w and the area of the class region. The size formula of the sampling area of each class is:

[0023]

[0024] Perform the loss function calculation process that is beneficial to improving the segmentation accuracy:

[0025] Introduce local contrast loss in the atrous spatial pyramid pooling module. By jointly using cross-entropy loss and DICE loss, the key weight coefficients α = 0.23, β = 0.36, γ = 0.41, τ = 0.07 are obtained in the experiment. According to the key weight coefficients α = 0.23, β = 0.36, γ = 0.41, τ = 0.07, the loss function is obtained. The loss function formula is:

[0026]

[0027]

[0028] In the above formula, is the total loss function, is the cross-entropy loss function, is the DICE loss function; is the local contrast loss function; τ is a hyperparameter that controls the smoothness of the feature similarity distribution in contrast learning; N represents the total number of pixels in the image, C represents the number of categories, including the number of background, the number of asphalt, and the number of aggregates, Y i,c indicates whether pixel i belongs to category c, p i,c represents the predicted probability that pixel i belongs to category c; p i represents the predicted probability of the model for pixel i, Y i represents the actual label of pixel i; z i and z j are the positive sample pairs of the feature vectors, z k is the negative sample feature vector, sim(z i ,z j ) is the cosine similarity;

[0029] Through the calculation of the loss function, the quantitative evaluation conclusion of the adhesion performance of aggregate asphalt is obtained. When the task and data scale are determined, when is lower than 0.2, it indicates that the model is in a good fitting state, indicating that the asphalt coverage rate is in a good state, corresponding to good adhesion performance of aggregate asphalt; when is still greater than 1.0 after twenty rounds of iterative training, it usually means that the model has not converged or is underfitting, indicating that the asphalt coverage rate is in a poor state, corresponding to poor adhesion performance of aggregate asphalt. Subsequently, it is necessary to correspondingly increase the number of training rounds or adjust the network structure and hyperparameters, and then re-obtain the basic data for calculation.

[0030] As a preferred solution: The process of quantitatively evaluating the aggregate asphalt adhesion performance of the basic data also includes the calculation processes of the inter-class separability index and the intra-class aggregation index. The specific calculation process is as follows:

[0031] Inter-class separability index

[0032]

[0033] Intra-class aggregation index

[0034]

[0035] In the above formula, and are the centroids of classes c1 and c2 respectively, and x i is the feature vector of the i-th sample in the class;

[0036] Through multiple validations of the model, the threshold of the inter-class separability index is obtained as 2.0, and the threshold of the intra-class aggregation index is 0.8; when the inter-class separability index is greater than 2.0 and at the same time the intra-class aggregation index is less than 0.8, it indicates that the features of different classes have a high degree of differentiation in space, and the internal feature distribution of the same class is in a compact state, which is conducive to the accurate prediction process of the residual asphalt coverage rate by the model.

[0037] The beneficial effects of the present invention are as follows:

[0038] 1. In the present invention, the aggregate asphalt adhesion rotary tester can simulate the dynamic action under actual road conditions through the mutual cooperation of the bottom plate, the outer shell and the test component, so as to realize the accurate evaluation of the adhesion form of the aggregate and the asphalt. At the same time, it can also dynamically simulate different working conditions, which is conducive to the subsequent accurate test of the peeling degree of the asphalt on the aggregate surface, and provides favorable image data conditions for the subsequent continuous and accurate evaluation of the coverage rate of the aggregate asphalt.

[0039] 2. The aggregate asphalt adhesion rotary tester in the present invention has good repeatability, controllable test conditions, wide application range, simple and fast, which is conducive to accurately and quickly obtaining a large number of concentrated image data, realizing large-scale evaluation, and is conducive to obtaining comprehensive and reliable subsequent evaluation results of the aggregate asphalt coverage rate and adhesion performance.

[0040] 3. The evaluation standard formed by the aggregate asphalt adhesion performance evaluation method in the present invention also perfects the defects of the international rotating bottle method. It realizes the dynamic simulation and accurate evaluation of the adhesion between the aggregate and the asphalt. Through the rotating bottle test, the peeling degree of the asphalt on the aggregate surface is measured, overcoming the shortcomings of the traditional boiling method and water immersion method, and providing a more scientific, simple and reliable test method, which is suitable for popularization and use. Description of the Drawings

[0041] Figure 1 Schematic diagram of the three-dimensional structure of the present invention;

[0042] Figure 2 Schematic side view of the connection relationship between the bar-shaped track, support frame, motor, gear and light source;

[0043] Figure 3 Schematic side view of the connection relationship between the first support plate, chain, transition gear, passive gear, active gear, driving member, specimen test bottle and bar-shaped support plate, only showing part of the chain in the figure;

[0044] Figure 4 Schematic three-dimensional view of the connection relationship between the bottom plate, first support plate, second support plate, roller body, transmission sprocket assembly, driving member, specimen test bottle and bar-shaped support plate;

[0045] Figure 5 Schematic top view of the connection relationship between the main support plate, bar-shaped track, self-moving light source, embedded camera and shadowless array light source bar;

[0046] Figure 6 Schematic side view of the connection relationship between the main support plate, bar-shaped track, support frame, motor, gear, light source and polyester reflective film sheet;

[0047] Figure 7 Schematic top view of the connection relationship between the bottom plate, first support plate, roller body, transmission sprocket assembly, driving member, bar-shaped support plate and paving plate;

[0048] Figure 8 Schematic top view of the connection relationship between the roller body and the specimen test bottle;

[0049] Figure 9 Schematic side view of the connection relationship between the bottom plate, first support plate, second support plate, roller body, specimen test bottle and support feet;

[0050] Figure 10 Schematic side view of the connection relationship between the bottom plate, outer shell, support feet and handle;

[0051] Figure 11 Operation flowchart of the method for evaluating the adhesion performance of aggregate asphalt in the present invention.

[0052] In the figure: 1 - bottom plate; 2 - outer shell; 3 - first support plate; 4 - second support plate; 5 - roller body; 6 - transmission sprocket assembly; 6-1 - chain; 6-2 - transition gear; 6-3 - passive gear; 6-4 - driving gear; 7 - driving member; 8 - clamping gap; 9 - multi-position irradiation shadowless lamp assembly; 9-1 - main support plate; 9-2 - strip track; 9-3 - self-moving light source; 9-3-1 - support frame; 9-3-2 - motor; 9-3-3 - gear; 9-3-4 - light source; 10 - specimen test bottle; 11 - first opening; 12 - opening and closing cover plate; 13 - second opening; 14 - viewing window; 15 - strip support plate; 16 - spreading plate; 17 - polyester reflective film sheet; 18 - embedded camera; 19 - shadowless array light source strip; 20 - support leg; 21 - heat dissipation hole; 22 - handle. Detailed implementation mode

[0053] The following uses specific specific examples to illustrate the implementation mode of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific implementation modes. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0054] Detailed implementation mode one: Combine Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 , Figure 9 and Figure 10 to illustrate this implementation mode. The aggregate asphalt adhesion rotation tester of this implementation mode includes a bottom plate 1, an outer shell 2 and a test component. The test component is arranged on the bottom plate 1, the outer shell 2 is arranged on the bottom plate 1, and an inner cavity is formed between the outer shell 2 and the bottom plate 1. The test component is arranged in the inner cavity;

[0055] The test component includes a first support plate 3, a second support plate 4, a plurality of roller bodies 5, a transmission sprocket assembly 6 and a driving member 7. The first support plate 3 and the second support plate 4 are arranged vertically and side by side on the bottom plate 1. The transmission sprocket assembly 6 is arranged on the first support plate 3. The transmission sprocket assembly 6 is connected to the driving member 7. A plurality of roller bodies 5 are arranged between the first support plate 3 and the second support plate 4. One end of each roller body 5 passes through the first support plate 3 and is connected to the transmission sprocket assembly 6. The other end of each roller body 5 is hinged to the second support plate 4. Each roller body 5 makes a self-rotation movement between the first support plate 3 and the second support plate 4 under the drive of the transmission sprocket assembly 6. A clamping gap 8 for cooperating with the specimen test bottle 10 is formed between every two adjacent roller bodies 5.

[0056] Further, an outer shell 2 and a test component are respectively arranged above the bottom plate 1, and four supporting feet 20 are arranged below the bottom plate 1. The four supporting feet 20 are arranged near the four angles of the bottom plate 1. The supporting feet 20 provide stable support for the rotary tester. The test component is arranged in the outer shell 2. A plurality of heat dissipation holes 21 are respectively machined on both sides of the outer shell 2 along its thickness direction. The heat dissipation holes 21 can effectively discharge the heat generated by the test component during operation, ensuring that the temperature of the rotary tester is stable, thereby avoiding the influence of temperature change on the test results. The specimen test bottle 10 is a bottle body containing asphalt and aggregates. The specimen test bottle 10 is arranged in the clamping gap 8 of each roller body 5. Each roller body 5 realizes a self-rotation movement driven by the transmission sprocket assembly 6, so as to ensure that the specimen test bottle 10 can realize the function of continuous rotation, and ensure that the asphalt and aggregates in the specimen test bottle 10 can be evenly mixed together during the rotation process.

[0057] Specific Embodiment 2: This embodiment is a further limitation of Specific Embodiment 1. A first opening 11 is machined on the outer shell 2, and an opening and closing cover plate 12 is hinged on the first opening 11.

[0058] Further, a handle 22 is arranged on the outer shell 2. The handle 22 is in a U shape. One side of the handle 22 is connected to the outer shell 2, and the other side of the handle 22 is suspended. The staff can open or close the opening and closing cover plate 12 through the handle 22.

[0059] Specific Embodiment 3: This embodiment is a further limitation of Specific Embodiment 1 or 2. A second opening 13 is machined on the outer shell 2, and a viewing window 14 is arranged on the second opening 13.

[0060] Further, the staff uses the handle 22 to open the opening and closing cover plate 12, and places the specimen test bottle 10 between two adjacent roller bodies 5, and then closes the opening and closing cover plate 12. The staff observes the test results through the viewing window 14.

[0061] Specific Embodiment 4: This embodiment is a further limitation of Specific Embodiment 1, 2 or 3. The transmission sprocket assembly 6 includes a chain 6-1, an intermediate gear 6-2, a driving gear 6-4 and a plurality of driven gears 6-3. The driven gears 6-3 are arranged in one-to-one correspondence with the roller bodies 5. Each driven gear 6-3 is sleeved on its corresponding roller body 5. One end of the roller body 5 passes through the first support plate 3 and is connected to its corresponding driven gear 6-3. The intermediate gear 6-2 is arranged on the outer wall of the first support plate 3. The driving gear 6-4 is sleeved on the power output shaft of the driving member 7. A chain 6-1 is arranged between the driving gear 6-4, the intermediate gear 6-2 and the plurality of driven gears 6-3.

[0062] Further, the roller body 5 is horizontally arranged between the first support plate 3 and the second support plate 4. One end of the roller body 5 passes through the first support plate 3 and is connected to the passive gear 6-3. At the same time, a transition gear 6-2 is arranged on the outer wall of the first support plate 3. The driving member 7 has the same working principle as the motor in the prior art. The driving member 7 can drive the driving gear 6-4 to rotate. The driving gear 6-4 is matched with the transition gear 6-2 and multiple passive gears 6-3 through the chain 6-1 to ensure that the roller body 5 realizes rotational motion. Heat is easily generated during the operation of the driving member 7, and the heat can be discharged through the heat dissipation holes 21 on the housing 2 to ensure the temperature stability inside the rotary tester.

[0063] Specific Embodiment 5: This embodiment is a further limitation of Specific Embodiment 1, 2, 3 or 4. A strip-shaped support plate 15 is arranged on the outer wall of the first support plate 3 along its length direction, and the chain 6-1 is arranged on the strip-shaped support plate 15.

[0064] Further, the chain 6-1 meshes with the transition gear 6-2, the driving gear 6-4 and multiple passive gears 6-3 respectively. The chain 6-1 is arranged on the strip-shaped support plate 15, and the strip-shaped support plate 15 ensures the stable operation of the chain 6-1 and avoids the phenomenon of falling off. The chain 6-1 has the same working principle as the chain in the prior art.

[0065] Specific Embodiment 6: This embodiment is a further limitation of Specific Embodiment 1, 2, 3, 4 or 5. It further includes a multi-position irradiation shadowless lamp assembly 9. One arrangement form of the multi-position irradiation shadowless lamp assembly 9 is that the multi-position irradiation shadowless lamp assembly 9 is arranged on the inner wall of the housing 2. The multi-position irradiation shadowless lamp assembly 9 is cooperatively provided with a paving plate 16. The paving plate 16 is arranged directly below the multi-position irradiation shadowless lamp assembly 9. The paving plate 16 is arranged on the bottom plate 1. The multi-position irradiation shadowless lamp assembly 9 includes a main support plate 9-1, a plurality of strip-shaped tracks 9-2 and a plurality of self-moving light sources 9-3. A plurality of strip-shaped tracks 9-2 are arranged on the main support plate 9-1, and a plurality of self-moving light sources 9-3 are arranged on each strip-shaped track 9-2; each self-moving light source 9-3 includes a support frame 9-3-1, a motor 9-3-2, a gear 9-3-3 and a light source 9-3-4. The support frame 9-3-1 is arranged on its corresponding strip-shaped track 9-2. One side of the support frame 9-3-1 is slidably matched with one side of the strip-shaped track 9-2. A motor 9-3-2 is arranged on the other side of the support frame 9-3-1. A gear 9-3-3 is sleeved on the output shaft of the motor 9-3-2. The gear 9-3-3 meshes with the strip-shaped track 9-2. A light source 9-3-4 is hinged to the bottom of the support frame 9-3-1. The light emitting direction of the light source 9-3-4 is directly downward or obliquely downward. The light source 9-3-4 reciprocally moves along the length direction of the strip-shaped track 9-2 under the drive of the support frame 9-3-1.

[0066] Further, the multi - illumination shadowless lamp assembly 9 further includes a plurality of shadowless array light source bars 19. The plurality of shadowless array light source bars 19 are respectively arranged near both sides and both ends of the main support plate 9 - 1. Each strip - shaped track 9 - 2 is arranged in the shadowless array light source bar 19, and a number of self - moving light sources 9 - 3 are arranged on each strip - shaped track 9 - 2. The motor 9 - 3 - 2 drives the support frame 9 - 3 - 1 to reciprocate on the strip - shaped track 9 - 2, so as to ensure that the light source 9 - 3 - 4 on the support frame 9 - 3 - 1 can reciprocate. The motor 9 - 3 - 2 has the same working principle as the motors in the prior art.

[0067] In this embodiment, a plurality of support blocks are arranged on the side of the main support plate 9 - 1 facing the bottom plate 1. The support blocks are vertically arranged at both ends of the strip - shaped track 9 - 2, so as to ensure that a number of self - moving light sources 9 - 3 reciprocate and slide on the strip - shaped track 9 - 2.

[0068] The configuration of the multi - illumination shadowless lamp assembly 9 in this embodiment is conducive to accurately obtaining the basic data in the adhesion evaluation method. Its significance lies in eliminating the interference of the external environment on the captured images and providing a standardized picture - taking environment. Among them, a number of self - moving light sources 9 - 3 and a plurality of shadowless array light source bars 19 cooperate with each other to form a uniform light source array structure. The plurality of shadowless array light source bars 19 can also be provided with a lifting mechanism. The lifting mechanism is used to lift or lower the position of the shadowless array light source bars 19. A lifting mechanism formed by the cooperation of a gear and a rack can be used, and other existing lifting mechanisms can also be replaced. The initial installation height of the shadowless array light source bars 19 is set to 50 cm, and the light source height is allowed to be adjusted between 30 cm and 60 cm to adapt to shooting objects of different sizes.

[0069] Further, the specific parameters of the shadowless array light source bar 19 are as follows:

[0070] Luminous flux: The luminous flux of each shadowless array light source bar 19 should be between 800 and 1200 lumens (lm) to ensure sufficient lighting effect.

[0071] Color temperature: The color temperature of the shadowless array light source bar 19 is set to 5000K to 6500K, approaching the color temperature of sunlight, so as to provide a natural lighting effect, which is a suitable shooting condition for aggregates.

[0072] Light source type: The shadowless array light source bar 19 is an LED light source, so as to provide an efficient and low - heat lighting solution and prevent overheating from affecting the normal use performance of aggregates and the embedded camera 18.

[0073] Front - to - back light source tilt angle: Set to 15 degrees, and the irradiation direction is towards the center; the left - to - right light source is also tilted 15 degrees to ensure the cross - illumination effect.

[0074] In this embodiment, the specific form of the cooperation among the main support plate 9-1, the strip track 9-2, and the self-moving light source 9-3 is as follows: the self-moving light source 9-3 can adjust the position and angle of the light source in real time according to different shooting requirements. The setting of the self-moving light source 9-3 ensures flexible adjustment of the lighting conditions during shooting to adapt to the characteristics of the aggregate, thereby obtaining the best lighting effect. The self-moving light source 9-3 system better adapts to the reflection characteristics of different materials, further improving the image quality and data availability. The self-moving light source 9-3 is fixed on the main support plate 9-1 through the strip track 9-2. There are a total of 8 strip tracks 9-2, and each strip track 9-2 can install multiple self-moving light sources 9-3. Specifically in use, the relevant parameters of the light source 9-3-4 in the self-moving light source 9-3 are as follows: the self-moving light source 9-3 has six degrees of freedom, the adjustment angle range is -30 degrees to +30 degrees, and it supports a 360-degree rotation function. The control system is adjusted by wireless remote control and integrates a light intensity sensor to monitor the lighting effect in real time. The light source type is LED, the luminous flux is set between 800 and 1200 lumens, and the color temperature range is 5000K to 6500K.

[0075] In this embodiment, the specific situation during the laying process of the polyester reflective film sheet 17 is as follows: the polyester reflective film sheet 17 is used as a shadowless structure. Specifically, a high-quality polypropylene scattering coating is formed on the inner surface of the housing 2 to form the polyester reflective film sheet 17, so that the reflectivity of the polyester reflective film sheet 17 should not be less than 85%, and the coating thickness is 5 mm. This material can effectively diffuse the incident light, ensuring that it evenly irradiates the entire shooting area, thereby significantly reducing the generation of local high-brightness areas and shadows, and improving the overall uniformity and quality of the image. The color of the bottom plate 1 is green, and the specific RGB value is (0, 255, 0). Through experimental verification, the green background performs best in the RGB color space of subsequent software processing, enhancing the image recognition effect.

[0076] In this embodiment, the embedded camera 18 can achieve standardized shooting. The specific situation is as follows: the embedded camera 18 is inlaid and connected with the main support plate 9-1, enabling the embedded camera 18 to adapt to different types of aggregates. This design significantly simplifies the operation process and reduces human errors, thereby improving the unified standardization effect of shooting. According to the different characteristics of the aggregates, the embedded camera 18 will automatically configure the focal length, aperture, ISO, white balance, and exposure time to ensure the best image quality in various environments. The shooting form achieved by the embedded camera 18 not only improves the accuracy of data collection but also provides strong support for subsequent image recognition and analysis, providing favorable support for the comprehensive evaluation of the adhesion between aggregates and asphalt.

[0077] The relevant setting parameters of the embedded camera 18 are:

[0078] When the shutter speed of the embedded camera 18 is 1 / 80 s, the corresponding ISO sensitivity is ISO 400, the focal length is 24 mm, and the aperture is f / 1.6;

[0079] When the shutter speed of the embedded camera 18 is 1 / 60 s, the corresponding ISO sensitivity is ISO 200, the focal length is 20 mm, and the aperture is f / 1.8;

[0080] When the shutter speed of the embedded camera 18 is 1 / 90 s, the corresponding ISO sensitivity is ISO 300, the focal length is 24 mm, and the aperture is f / 1.5;

[0081] The above setting parameter configuration method is selected according to different situations and is adapted to comprehensively obtain effective and stable basic image data.

[0082] Specific Embodiment Seven: This embodiment is a further limitation of Specific Embodiments One, Two, Three, Four, Five, or Six, and further includes a multi-position illumination shadowless lamp assembly 9. Another arrangement form of the multi-position illumination shadowless lamp assembly 9 is that the multi-position illumination shadowless lamp assembly 9 is arranged outside the housing 2, and cooperates with the machine case and another paving board 16 to realize the independent acquisition process of a large number of shadowless images without relying on the bottom plate 1 and the housing 2, which is beneficial to accurately and effectively obtain basic data. The multi-position illumination shadowless lamp assembly 9 includes a main support plate 9-1, a plurality of strip-shaped tracks 9-2, and a plurality of self-moving light sources 9-3. A plurality of strip-shaped tracks 9-2 are arranged at the bottom of the main support plate 9-1, and a plurality of self-moving light sources 9-3 are arranged on each strip-shaped track 9-2; each self-moving light source 9-3 includes a support frame 9-3-1, a motor 9-3-2, a gear 9-3-3, and a light source 9-3-4. The support frame 9-3-1 is arranged on its corresponding strip-shaped track 9-2. One side of the support frame 9-3-1 is slidably matched with one side of the strip-shaped track 9-2. A motor 9-3-2 is arranged on the other side of the support frame 9-3-1. A gear 9-3-3 is sleeved on the output shaft of the motor 9-3-2, and the gear 9-3-3 meshes with the strip-shaped track 9-2. A light source 9-3-4 is hinged to the bottom of the support frame 9-3-1, and the light emitting direction of the light source 9-3-4 is downward. The light source 9-3-4 reciprocates along the length direction of the strip-shaped track 9-2 under the drive of the support frame 9-3-1.

[0083] Specific Embodiment Eight: This embodiment is a further limitation of Specific Embodiments One, Two, Three, Four, Five, Six, or Seven. A polyester reflective film sheet 17 is arranged at the bottom of the main support plate 9-1, and a plurality of strip-shaped tracks 9-2 are arranged in an array form on the polyester reflective film sheet 17. An embedded camera 18 is arranged at the center of the bottom of the main support plate 9-1.

[0084] Further, the light source 9-3-4 and the shadowless array light source bar 19 illuminate the paving plate 16 and can be reflected on the polyester reflective film 17, thereby ensuring sufficient light inside the rotary tester. When the embedded camera 18 takes pictures, it ensures the high quality of the captured images, and the finally captured photos will be used for later image recognition processing, providing a reliable data basis for image recognition.

[0085] In the present invention, the installation process and working principle of the aggregate asphalt adhesion rotary tester: The outer shell 2 is installed on the bottom plate 1, and the test component is installed inside the outer shell 2. Among them, the first support plate 3 and the second support plate 4 on the test component are vertically installed on the bottom plate 1. One end of multiple rollers 5 passes through the first support plate 3 and is connected to the transmission sprocket assembly 6, and the other end of multiple rollers 5 is hinged on the second support plate 4. Start the driving member 7, and the driving member 7 drives multiple rollers 5 to rotate on the first support plate 3 and the second support plate 4 through the transmission sprocket assembly 6. The staff places the specimen test bottle 10 in the clamping gap 8 between two rollers 5, and the specimen test bottle 10 realizes a rotational motion.

[0086] Specific Embodiment Nine: Combine Figures 1 to 11 To illustrate this embodiment, in this embodiment, the aggregate asphalt adhesion rotary tester includes a bottom plate 1, an outer shell 2, and a test component. The test component is arranged on the bottom plate 1, the outer shell 2 is arranged on the bottom plate 1, and an inner cavity is formed between the outer shell 2 and the bottom plate 1. The test component is arranged in the inner cavity;

[0087] The test component includes a first support plate 3, a second support plate 4, multiple rollers 5, a transmission sprocket assembly 6, and a driving member 7. The first support plate 3 and the second support plate 4 are vertically arranged side by side on the bottom plate 1. The transmission sprocket assembly 6 is arranged on the first support plate 3. The transmission sprocket assembly 6 is connected to the driving member 7. Multiple rollers 5 are arranged between the first support plate 3 and the second support plate 4. One end of each roller 5 passes through the first support plate 3 and is connected to the transmission sprocket assembly 6. The other end of each roller 5 is hinged to the second support plate 4. Each roller 5 makes a self-rotation motion between the first support plate 3 and the second support plate 4 driven by the transmission sprocket assembly 6. A clamping gap 8 for cooperating with the specimen test bottle 10 is formed between every two adjacent rollers 5.

[0088] In this embodiment, the method for evaluating the aggregate asphalt adhesion performance realized by the aggregate asphalt adhesion rotary tester is to use the image data collected by the aggregate asphalt adhesion rotary tester as the basic data, model the basic data to form a relevant model, and thus complete the quantitative evaluation process of the aggregate asphalt adhesion performance according to the model. That is, using the image data collected by the aggregate asphalt adhesion rotary tester as the basic data, and performing a quantitative evaluation on the basic data through an image recognition method based on deep learning. Specifically:

[0089] The process of quantitative evaluation through basic data is to form a training set from the image data obtained by the aggregate asphalt adhesion rotary tester. After randomly cropping a predetermined number of images from the training set, classification is performed. The class center point of each category is defined as the arithmetic mean of the pixel coordinates of all pixels in that category. The calculation formula for the arithmetic mean is:

[0090]

[0091] In the above formula, (x i , y i ) is the coordinate of the i-th pixel point in category C, and N C is the number of pixels in category C; then a weight coefficient w is introduced for each category, and the sampling probability of each category is adjusted according to the weight coefficient w. The sampling probability P of each category is calculated according to its corresponding weight coefficient w. The calculation formula for the sampling probability P is:

[0092]

[0093] According to the formula for the sampling area of each category, the size of the sampling area is calculated through the weight coefficient w and the area of the category region. The formula for the size of the sampling area of each category is:

[0094]

[0095] Perform the calculation process of the loss function that is beneficial to improving the segmentation accuracy:

[0096] In the atrous spatial pyramid pooling module, local contrast loss is introduced. By jointly using cross-entropy loss and DICE loss, the key weight coefficients α = 0.23, β = 0.36, γ = 0.41, τ = 0.07 are obtained in the experiment. According to the key weight coefficients α = 0.23, β = 0.36, γ = 0.41, τ = 0.07, the loss function is obtained. The formula for the loss function is:

[0097]

[0098] In the above formula, is the total loss function, is the cross-entropy loss function, is the DICE loss function, is the local contrast loss function. The total loss function combines the cross-entropy loss function the DICE loss function and the local contrast loss function to realize the optimization process of the entire model; the cross-entropy loss function Used to measure the difference between the predicted probability distribution and the true label distribution. DICE loss function Is the DICE loss, which is used to measure the quality of the overlap between the model prediction and the true label region. Local contrast loss function The main purpose is to make the feature vectors of samples in the same category more similar and the feature vectors of different categories more distinguishable by optimizing the sample distribution in the feature space. τ is an important hyperparameter in contrastive learning, which is used to control the smoothness of the feature similarity distribution; τ is a hyperparameter that controls the smoothness of the feature similarity distribution in contrastive learning; N represents the total number of pixels in the image, C represents the number of categories, including the number of backgrounds, the number of asphalt, and the number of aggregates, Y i,c Indicates whether pixel i belongs to category c, p i,c Represents the predicted probability that pixel i belongs to category c; p i Represents the predicted probability of the model for pixel i, Y i Indicates the actual label of pixel i; z i And z j Are the positive sample pairs of the feature vectors, z k Is the negative sample feature vector, sim(z i ,z j ) is the cosine similarity;

[0099] Through the calculation of the loss function, the quantitative evaluation conclusion of the adhesion performance of aggregate asphalt is that when the task and data scale are determined, when Is lower than 0.2, it indicates that the model is in a good fitting state, indicating that the asphalt coverage is in a good state, corresponding to good aggregate asphalt adhesion performance; when Is still greater than 1.0 after twenty rounds of iterative training, which usually means that the model has not converged or is underfitting, indicating that the asphalt coverage is in a poor state, corresponding to poor aggregate asphalt adhesion performance. Subsequently, it is necessary to increase the number of training rounds or adjust the network structure and hyperparameters and then re-obtain the basic data for calculation.

[0100] Specific implementation method ten: This implementation method is a further limitation of the ninth specific implementation method. In the actual training process, It will converge to a relatively stable numerical range, specifically depending on the different limitations of the task and data scale. The total loss function Generally, being lower than 0.2 can be regarded as a good fit; if it is greater than 1.0 for a long time, it usually means that the model has not converged or is underfitting, and it is necessary to increase the number of training rounds or adjust the network structure, hyperparameters, etc. When the In this model converges to 0.147, it indicates that it has an excellent good fitting effect and can accurately predict the asphalt coverage.

[0101] The aggregate asphalt adhesion performance evaluation method in the present invention can accurately measure the asphalt coverage rate and further evaluate the adhesion performance based on this. After a large number of research and calculations, it is summarized as follows:

[0102] When the coverage rate is greater than or equal to 90%, the adhesion performance is usually considered very good; when the coverage rate is between 80% and 90%, the adhesion performance can be regarded as good; when the coverage rate is between 70% and 80%, it indicates that the adhesion performance is at a medium level; and when the coverage rate is less than 70%, the adhesion performance is usually considered poor.

[0103] Specific Embodiment XI: This embodiment is a further limitation of Specific Embodiment IX or X. In this embodiment, the aggregate asphalt adhesion performance evaluation method for quantitatively evaluating the aggregate asphalt adhesion performance from the basic data further includes the calculation processes of the inter-class separability index and the intra-class aggregation index. The specific calculation processes are as follows:

[0104] Inter-class separability index

[0105]

[0106] Intra-class aggregation index

[0107]

[0108] In the above formula, and are the centroids of classes c1 and c2 respectively, and x i is the feature vector of the i-th sample in the class.

[0109] Similar to the loss function, the main function of the inter-class separability index is to evaluate the accuracy of the evaluation results in the aggregate asphalt adhesion performance evaluation method. Through multiple validations of the model, the threshold of the inter-class separability index is obtained as 2.0, and the threshold of the intra-class aggregation index is 0.8; when this index is greater than 2.0, it indicates that the features of different classes have a high degree of distinguishability in space; at the same time, the threshold of the intra-class aggregation index is set to 0.8: when this index is less than 0.8, it shows that the internal feature distribution of the same class is more compact, so that higher recognition accuracy and stability can be obtained in practical applications. Under this threshold, the model can accurately predict the residual asphalt coverage rate.

[0110] Embodiment Twelve: This embodiment is a further limitation of Embodiment Nine, Ten or Eleven. In this embodiment, the multi-position illumination shadowless lamp assembly 9 is used to spread the aggregate and asphalt mixture after the rotary bottle test on the paving plate 16 after the rotary bottle test. The multi-position illumination shadowless lamp assembly 9 irradiates and collects relevant images of the aggregate and asphalt mixture on the paving plate 16, so as to form basic data for studying the adhesion condition of the aggregate and asphalt, so as to perform image recognition and analysis subsequently. The inner cavity provides an independent and controllable shooting environment to avoid the interference of external light and background. The cooperation between the shadowless array light source strip 19 and the multiple self-moving light sources 9-3 realizes the layout structure of the dynamic and static circumferential light sources, which can ensure uniform light during shooting and eliminate the influence of shadows. The self-moving light source 9-3 can adjust the light source position and angle according to different shooting requirements to enhance the adaptability of the light. The viewing window 14 is convenient for observing and adjusting the shooting angle.

[0111] The specific operation process of this aggregate asphalt adhesion rotary tester is as follows:

[0112] First step: After the prepared aggregate and asphalt mixture specimens are fully cooled, the staff puts them into the aggregate asphalt bottle 10 to ensure that the aggregate is evenly coated with asphalt and is fixed on the multiple rollers 5. A plurality of rollers 5 are installed between the second support plate 4 and the first support plate 3, and the aggregate asphalt bottle 10 is installed between every two adjacent rollers 5 to ensure the stability of the multiple rollers 5 driving the aggregate asphalt bottle 10 during rotation.

[0113] Second step: Set the rotation speed V and rotation time T of the aggregate asphalt bottle 10 through the control system. Set the rotation speed required for the test through the rotation speed control button. The rotation speed and time should be adjusted according to the test requirements and specifications to simulate different working conditions.

[0114] Third step: Start the driving member 7, and the driving member 7 drives the driving gear 6-4 to perform a self-rotation movement. The driving gear 6-4 drives the intermediate gear 6-2 and a plurality of driven gears 6-3 to rotate through the chain 6-1, ensuring that the rollers 5 rotate and transmit power to the aggregate asphalt bottle 10, ensuring that the aggregate asphalt bottle 10 rotates at the set rotation speed. The aggregate and asphalt mixture moves in the water medium environment in the aggregate asphalt bottle 10 to simulate the working conditions of asphalt and aggregate in the actual road environment.

[0115] Fourth step: Through the control system and the display screen, the rotation speed and time during the operation of the device are monitored and recorded in real time, and the peeling situation of the asphalt in the aggregate asphalt bottle 10 is observed through the viewing window 14. After the specified rotation time, the asphalt mixture is taken out, and the peeling degree of the asphalt on the aggregate surface is observed to evaluate the asphalt-aggregate adhesion.

[0116] Step 5: After the test of the aggregate asphalt bottle 10, the asphalt mixture is moved to the paving plate 16 and photographed by the embedded camera 18 on the multi - illumination shadowless lamp assembly 9. The housing 2 provides an independent shooting environment. During shooting, the embedded camera 18 is illuminated by the multi - illumination shadowless lamp assembly 9 to eliminate shadow interference and ensure the clarity and consistency of the images. The multi - illumination shadowless lamp assembly 9 ensures that the object to be photographed is in the best position, thus obtaining high - quality images. These images will be used for subsequent analysis, and the deep - learning - based method proposed in the present invention is adopted to further evaluate the adhesion of asphalt - aggregate.

[0117] The present invention proposes a standard form of image acquisition, standardizes the acquisition process, avoids interference caused by external factors during image acquisition, completely eliminates shadow interference through the multi - illumination shadowless lamp assembly 9, and at the same time forms quantitative shooting distances, the material and color of the bottom plate 1, and the parameters of picture shooting, improving the standardized quality of image acquisition.

[0118] The present invention develops an image recognition software based on computer vision technology and deep - learning technology. The present invention provides a pre - processing scheme for the collected pictures, collects and annotates a large - scale data set, and solves the problem that there is no similar data set in the current related fields. At the same time, the present invention uses this data set to train a semantic segmentation model with ResNet - 101 as the backbone network, providing a high - precision automatic recognition scheme. It solves the problems of low precision, high error, and complexity of the previous evaluation methods.

[0119] The adhesion evaluation method in the present invention applies deep - learning technology to the evaluation process of the rotating bottle test, breaking through the traditional method based on threshold segmentation. Compared with the common manual annotation and empirical judgment in the industry, for the first time, a deep - learning semantic segmentation model with ResNet - 101 as the backbone network is proposed and introduced to realize the automatic processing of images in the rolling bottle test. This adhesion evaluation method optimizes the evaluation process through deep learning, solves the subjectivity problem caused by relying on manual experience in traditional methods, and improves the objectivity and accuracy of the evaluation results. The introduction of this technology provides a more accurate and efficient evaluation means for the rolling bottle test.

[0120] The adhesion evaluation method in the present invention is based on a semantic segmentation model with the ResNet-101 architecture, which has obvious advantages in solving the problem of class imbalance and improving segmentation accuracy. First, by combining ResNet-101 as the feature extraction backbone network, compared with traditional U-Net, ResNet-50 or other shallow networks, it can more effectively capture local information in complex images. Second, a self-created class center sampling method is proposed. By ensuring that samples of all classes are effectively sampled in each training cycle, especially by repeating sampling to increase the training probability of asphalt pixels, the problem of class imbalance is successfully alleviated, and the training process is accelerated. At the same time, local contrast loss and the ASPP atrous spatial pyramid pooling module are introduced, enabling the model to finely process image features at multiple scales, further improving the boundary processing accuracy and overall segmentation effect. A new loss function is introduced to further optimize the classification performance and effectively address the challenge of class imbalance. This adhesion evaluation method combines sampling strategy, loss function, and network architecture, not only improving the segmentation accuracy but also enhancing the generalization ability of the model.

[0121] Combined Figure 11 As shown, the adhesion evaluation method in the present invention also includes an image recognition algorithm, which adopts a centroid sampling-based strategy and effectively solves the problem of class imbalance between asphalt and aggregate categories in the training dataset. By randomly sampling the centroid of the category in each training cycle and cropping out image patches of a fixed size from the centroid point of the category, the recognition ability of the model on different categories is significantly enhanced. Especially when dealing with small-sample categories, this method greatly improves the generalization ability of the model.

[0122] During the experiment, the present invention randomly cropped 1200 images from the training set, each with a size of 600×600 pixels, and through verification, it was found that the image size of 600×600 pixels achieved the best balance between efficiency and accuracy. This size not only improves the training speed and efficiency but also solves the problem of pixel imbalance among the three types of residual asphalt, aggregate, and background while retaining sufficient detail information. By reasonably selecting the image size, the model can maintain high accuracy performance among different categories and avoid the influence brought by class imbalance. The selection of 600×600 pixels not only takes into account the optimization of computing resources but also improves the accuracy of the model.

[0123] The class center point of each category is defined as the arithmetic mean of the pixel coordinates of all pixels in that category. The specific formula is as follows:

[0124]

[0125] where, (x i ,y i) is the coordinate of the i-th pixel in category C, N C is the number of pixels in category C.

[0126] To further optimize the sampling process, the present invention introduces a weight coefficient w for each category and adjusts the sampling probability of each category according to these weight coefficients. The sampling probability P of each category is calculated based on its weight coefficient w. The sampling probability formula proposed by the present invention is as follows:

[0127]

[0128] In addition, the present invention also gives the area of the sampling region for each category The formula for calculating the size of the sampling region by the weight coefficient and the area of the category region. The formula for the size of the sampling region for each category proposed by the present invention is as follows:

[0129]

[0130] The present invention proposes a completely new and self-created loss function to improve the segmentation accuracy, specifically:

[0131] In the present invention, a local contrast loss method is introduced into the atrous spatial pyramid pooling module of the deep learning model to further optimize the discrimination between different categories in the feature space. This method effectively improves the segmentation accuracy of the model at the boundary details and enhances the recognition ability of the asphalt-aggregate interface region. It is used in combination with the cross-entropy loss and the DICE loss, and the key weight coefficients α = 0.23, β = 0.36, γ = 0.41, τ = 0.07 are obtained in a large number of experiments. The loss function proposed by the present invention is as shown in the following formula:

[0132]

[0133] Where: n represents the total number of pixels in the image, C represents the number of categories (such as background, asphalt, aggregate), Y i,c represents whether pixel i belongs to category c, p i,c represents the predicted probability that pixel i belongs to category c. p i represents the predicted probability of the model for pixel i, Y i represents the actual label of pixel i. z i and z j are the positive sample pairs of the feature vectors, z k is the negative sample feature vector, sim(z i ,z j ) is the cosine similarity.

[0134]

[0135]

[0136] Based on the experimental data, a set of optimal parameter values was proposed to optimize the Adam optimizer.

[0137] The present invention improves and optimizes the traditional Adam optimizer, adjusts the key parameters in the optimizer to better adapt to the specific task requirements. Through a large number of experiments and fitting of the experimental data, a set of optimal parameter values was obtained. These adjustments help to improve the convergence speed and accuracy of the model.

[0138] The specific formula of the optimized Adam optimizer is as follows:

[0139] First-order moment (momentum) update formula:

[0140] m t = 0.85m t-1 + 0.15g t

[0141] Second-order moment (momentum) update formula:

[0142] Bias correction formula:

[0143]

[0144] Parameter update formula:

[0145]

[0146] By making adaptive adjustments to these parameters, through the fitting optimization of a large amount of experimental data, the present invention solves the efficiency problem of the traditional Adam optimizer in certain specific tasks. The adjusted Adam optimizer can improve the model performance while accelerating the convergence process, avoiding the problems of slow convergence or instability in the traditional method.

[0147] In the present invention, the performance evaluation of the model not only depends on traditional metrics such as accuracy and recall, but also for the first time applies two evaluation metrics, Inter-class Separation and Intra-class Aggregation, to the evaluation of the adhesion between aggregate and asphalt. It can more deeply understand the performance of the model in the feature space and significantly optimize the classification effect of the adhesion characteristics between aggregate and asphalt. By introducing these two metrics, it can effectively evaluate the separation ability between different classes (residual asphalt, aggregate, and background) and the consistency of samples within the same class of the model, thereby improving the discrimination and precision of the model, especially under the complex conditions of the present invention.

[0148]

[0149] and are the centroids of classes c1 and c2 respectively, and x i is the feature vector of the i-th sample in the class.

[0150] For the first time, the present invention applies the separability between classes and the aggregation within classes to the algorithm for the adhesion between asphalt and aggregates, providing a brand-new perspective and technical path when the traditional method cannot effectively solve the problem. Through the adhesion evaluation method based on the deep learning algorithm, the present invention significantly improves the recognition ability of the model for the adhesion characteristics between aggregates and asphalt, providing favorable data support for the research in related fields.

Claims

1. A method for evaluating the adhesion performance of aggregate to asphalt, characterized in that: The method for evaluating the adhesion performance of aggregate asphalt is based on the image data collected by the aggregate asphalt adhesion rotary tester, and the process of quantitatively evaluating the aggregate asphalt adhesion of the basic data is as follows: The process of quantitative evaluation through the basic data is to form a training set from the image data obtained by the aggregate asphalt adhesion rotary tester. After randomly cropping a predetermined number of images from the training set to form categories, the class center point of each category is defined as the arithmetic mean of the pixel coordinates of all pixels in that category. The calculation formula for the arithmetic mean is: In the above formula, (x i , y i ) is the coordinate of the i-th pixel point in class C, and N C is the number of pixels in class C; Then, a weight coefficient w is introduced for each category, and the sampling probability of each category is adjusted according to the corresponding weight coefficient w. The sampling probability P of each category is calculated based on its corresponding weight coefficient w. The calculation formula for the sampling probability P is: According to the area of the sampling region for each category Based on the formula, the size of the sampling region is calculated through the weight coefficient w and the area of the category region. The size formula for the area of the sampling region for each category is as follows: Perform the calculation process of the loss function that is beneficial to improving the segmentation accuracy: A local contrast loss is introduced in the hole spatial pyramid pooling module. Through the joint use of the cross-entropy loss and the DICE loss, the key weight coefficients α = 0.23, β = 0.36, γ = 0.41, τ = 0.07 are obtained in the experiment. Based on the key weight coefficients α = 0.23, β = 0.36, γ = 0.41, τ = 0.07, the loss function is obtained. The loss function formula is: In the above formula, is the total loss function, is the cross-entropy loss function, is the DICE loss function; is the local contrast loss function; τ is a hyperparameter that controls the smoothness of the feature similarity distribution in contrastive learning; N represents the total number of pixels in the image, C represents the number of classes, including the number of background, the number of asphalt, and the number of aggregates, Y i,c indicates whether pixel i belongs to class c, p i,c represents the predicted probability that pixel i belongs to class c; p i represents the predicted probability of the model for pixel i, Y i represents the actual label of pixel i; z i and z j are the positive sample pairs of feature vectors, z k is the negative sample feature vector, sim(z i ,z j ) is the cosine similarity; Through the calculation of the loss function, the quantitative evaluation conclusion of the adhesion of aggregate to asphalt is obtained. When the task and data scale are determined, when is less than 0.2, it indicates that the model is in a good fitting state, indicating that the asphalt coverage rate is in a good state, corresponding to good aggregate asphalt adhesion performance; when is still greater than 1.0 after 20 rounds of iterative training, it usually means that the model has not converged or is underfitting, indicating that the asphalt coverage rate is in a poor state, corresponding to poor aggregate asphalt adhesion. Subsequently, it is necessary to increase the number of training rounds or adjust the network structure and hyperparameters accordingly, and then re-obtain the basic data for calculation.

2. The aggregate asphalt adhesion performance evaluation method according to claim 1, characterized in that: The process of quantitatively evaluating the aggregate asphalt adhesion of the basic data also includes the calculation processes of the inter-class separability index and the intra-class aggregation index. The specific calculation process is as follows: Inter-class separability index Intra-class aggregation index In the above formula, and are the centroids of categories c1 and c2 respectively, and x i is the feature vector of the i-th sample in the category; Through multiple validations of the model, the threshold of the inter-class separability index is obtained as 2.0, and the threshold of the intra-class aggregation index is 0.8; when the inter-class separability index is greater than 2.0 and at the same time the intra-class aggregation index is less than 0.8, it indicates that the features of different categories have a high degree of distinguishability in space, and the internal features of the same category are distributed in a compact state, which is conducive to the accurate prediction process of the model for the residual asphalt coverage rate.

Images