An interference fringe pattern positioning method based on target detection

Abstract

The application discloses an interference fringe pattern positioning method based on target detection. The application uses an interference particle imaging system to obtain an interference fringe pattern of a transparent spherical particle, divides the labeled interference fringe pattern into a training set, a verification set and a test set, obtains optimal weights after training for 200 epochs, takes the test set as input of a target detection network, obtains a prediction box containing normalized position information of the interference fringe pattern, and obtains a center point of the interference fringe pattern after coordinate extraction and inverse transformation. The application can realize high-precision positioning of the interference fringe pattern of the transparent spherical particle, and provides technical support for particle size measurement.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of image processing, and particularly relates to a method for obtaining position information of an interference fringe pattern using target detection. BACKGROUND

[0002] Clouds widely cover the earth's surface, which is composed of ice crystals and liquid droplets, and plays an important role in the earth's water cycle. In cloud and precipitation physics, according to the microstructure characteristics of the cloud, it can be divided into water cloud, ice cloud and mixed cloud. The acquisition of cloud microphysical information such as water content / ice content, particle size and shape in the cloud field is of great help to the study of cloud development process and precipitation formation mechanism and the realization of artificial rainfall. The shape of liquid droplet particles in the cloud can generally be regarded as spherical, and the position information of transparent spherical particles is of great importance to the measurement of particle size. Therefore, it is of great significance to study the high-precision interference fringe pattern positioning method.

[0003] The interference particle imaging technology has the advantages of wide measurement range, high accuracy, fast processing speed and non-contact. In recent years, the interference particle imaging technology has been widely applied in particle size measurement, phase state discrimination and other aspects. In the transparent spherical particle measurement using the interference particle imaging technology, the patent CN103674791A proposes an interference particle imaging measurement method based on double-beam irradiation, which obtains the center coordinates of the interference fringe pattern by using the one-way gradient matching method, and realizes the positioning of the interference fringe pattern. However, this method is not conducive to real-time detection.

[0004] Target detection is an important field in computer vision. Nowadays, target detection has been widely used in object positioning and classification tasks. Target detection network selects the region of interest for processing, detects the possible targets on the image and determines their positions and categories. The commonly used target detection networks include the One-Stage YOLO (You Only Look Once) series and the Two-Stage R-CNN (Regions with CNN features) series. The YOLO series network has good real-time performance and accuracy, and is very suitable for multi-scale target positioning. However, there is no example of using target detection method to position the transparent spherical particle interference fringe pattern in the prior art. The difficulties of using this method are as follows: first, a large amount of manually annotated data set is required. After obtaining the interference fringe pattern by using the interference particle imaging system, all targets in the interference fringe pattern must be accurately labeled. Second, the hyperparameters of the target detection network need to be adjusted to obtain the optimal weights. Including learning rate, cosine annealing parameter and weight decay coefficient. Third, the network structure needs to be modified to make the target detection network have stronger generalization ability. SUMMARY

[0005] The application aims to overcome the shortcomings of the prior art, and provide an interference fringe pattern positioning method based on target detection.

[0006] The technical scheme adopted by the application is as follows: an interference fringe pattern positioning method based on target detection is performed according to the following steps:

[0007] Step 1: build an interference particle imaging system, which includes a laser 1, a plano-convex cylindrical lens 5 and a plano-concave cylindrical lens 6; the laser 1, a microscope objective 2, a pinhole 3, a collimating lens 4, the plano-convex cylindrical lens 5, the plano-concave cylindrical lens 6 and a sample cell 7 are sequentially arranged in the horizontal direction from left to right; the sample cell 7 contains deionized water and polystyrene spherical particles; an imaging lens 8 and a CCD camera 9 are sequentially arranged above the sample cell 7; the laser 1 emits parallel polarized light; the laser beam passes through the spatial filter composed of the microscope objective 2 and the pinhole 3 to complete the beam expansion function; after passing through the collimating lens 4, the light beam becomes a parallel light beam; the two lens groups of the plano-convex cylindrical lens 5 and the plano-concave cylindrical lens 6 compress the light beam into a parallel light sheet; finally, the CCD camera is used to obtain an interference fringe pattern;

[0008] Step 2: after labeling the interference fringe pattern, randomly divide it into a training set, a validation set and a test set at a ratio of 6:2:2; set the network hyperparameters, put the training set into the target detection network for training for 200 epochs, and obtain the optimal weight;

[0009] Step 3: input the test set divided in step 2 into the YOLOv7 network to obtain a prediction box containing normalized position information;

[0010] Step 4: extract the position information in the prediction box to obtain the center point of the interference fringe pattern through coordinate inverse transformation; thus, the precise positioning function of the interference fringe pattern is realized.

[0011] The laser 1 emits parallel polarized light with a wavelength of 532 nm, and the maximum output power is 3 W.

[0012] A CCD camera with a resolution of 1920x1200 and a pixel size of 5.86um is used as a receiver, the system magnification is set to 0.86, and the system scattering angle is set to 90°.

[0013] The particles to be measured are 10um polystyrene spherical particles, which are placed in deionized water.

[0014] The application has the following technical effects:

[0015] First, real-time processing. The target is directly located by the prediction frame, and the center point coordinates of the interference fringe pattern are obtained through coordinate extraction and inverse transformation. Compared with the traditional image processing method, the method has higher real-time performance.

[0016] Second, strong generalization ability. The method can accurately detect the target position and has strong generalization ability when processing multi-scale and different fringe frequency interference fringe patterns obtained by different interference particle imaging systems.

[0017] Third, good robustness. Brightness change and noise interference are important factors affecting the quality of the interference fringe pattern. The method can accurately detect the target position when there are interference factors, and has good robustness.

[0018] The present application provides technical support for particle size measurement by high-precision positioning of the interference fringe pattern through the target detection method. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The flow chart of the transparent spherical particle interference fringe pattern positioning method of the present application is shown in the figure.

[0020] Figure 2 The structure of the interference particle imaging system is shown in the figure.

[0021] Figure 3 The positioning result of the interference fringe pattern is shown in the figure. DETAILED DESCRIPTION

[0022] The embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

[0023] The flow chart of the interference fringe pattern positioning method based on target detection of the present application is shown in the figure. Figure 1 Specifically,

[0024] Step 1: Obtain the interference fringe pattern of the transparent spherical particle: the interference particle imaging system used in the present application is shown in the figure. Figure 2The shown interference particle imaging system comprises a laser 1, a plano-convex cylindrical lens 5 and a plano-concave cylindrical lens 6; the laser 1, a microscope objective 2, a pinhole 3, a collimating lens 4, the plano-convex cylindrical lens 5, the plano-concave cylindrical lens 6, a sample cell 7 are sequentially arranged in the horizontal direction from left to right, the sample cell 7 contains deionized water and polystyrene spherical particles, and an imaging lens 8 and a CCD camera 9 are sequentially arranged above the sample cell 7; the laser 1 emits parallel polarized light with a wavelength of 532 nm, and the maximum output power is 3 W. The laser beam passes through the spatial filter composed of the microscope objective 2 and the pinhole 3 to complete the beam expansion function, and becomes a parallel light beam after passing through the collimating lens 4. The plano-convex cylindrical lens 5 and the plano-concave cylindrical lens 6 two lens groups compress the light beam into a parallel light sheet. A Nikon lens with a focal length of 50 mm is used as an imaging lens to collect particle scattering light. A CCD camera with a resolution of 1920x1200 and a pixel size of 5.86um is used as a receiving device. The system magnification is set to 0.86, and the system scattering angle is set to 90°. The particles to be measured are 10um polystyrene spherical particles, which are placed in deionized water.

[0025] Step 2: After labeling the interference fringe pattern, divide it into a training set, a validation set and a test set in a ratio of 6:2:2 at random; set the network hyperparameters: set the initial learning rate to 0.01, set the learning rate momentum to 0.937, set the weight decay coefficient to 0.0005, set the cosine annealing parameter to 0.1, set the bounding box loss weight coefficient to 0.05, set the anchor scaling threshold to 4.0, set the IoU training threshold to 0.2, and set the mosaic parameter to 1.0; put the training set into the target detection network for training for 200 epochs to obtain the optimal weight;

[0026] Step 3: input the test set divided in step 2 into the YOLOv7 network, and obtain the prediction box containing normalized position information after network detection;

[0027] Step 4: as shown in Figure 3 , extract the prediction box coordinates and perform inverse transformation to obtain the positioning information and draw it on the interference fringe pattern. Thus, the precise positioning function of the interference fringe pattern is realized.

Claims

1. A method for locating interferometric fringe patterns based on target detection, characterized in that, The method includes the following steps: Step (1): Use an interferometric particle imaging system to obtain the interference fringe pattern of transparent spherical particles; An interferometric particle imaging system was constructed, comprising a laser, a plano-convex cylindrical lens, and a plano-concave cylindrical lens. The laser, microscope objective, pinhole, collimating lens, plano-convex cylindrical lens, plano-concave cylindrical lens, and sample cell were arranged horizontally from left to right. The sample cell contained deionized water and polystyrene spherical particles. An imaging lens and a CCD camera were positioned above the sample cell. The system magnification was set to 0.86, and the system scattering angle was set to 90°. The laser emitted parallel polarized light. The laser beam was expanded by a spatial filter composed of the microscope objective and the pinhole. After passing through the collimating lens, it became a parallel beam. The plano-convex cylindrical lens and the plano-concave cylindrical lens compressed the beam into a parallel output light sheet. 10 μm polystyrene spherical particles were placed in the sample cell containing deionized water, and the interference fringe pattern was acquired using the CCD camera. Step (2): After labeling the interference fringe icons, randomly divide them into training, validation, and test sets in a 6:2:2 ratio; set the network hyperparameters: initial learning rate to 0.01, learning rate momentum to 0.937, weight decay coefficient to 0.0005, cosine annealing parameter to 0.1, bounding box loss weight coefficient to 0.05, anchor scaling threshold to 4.0, IoU training threshold to 0.2, and mosaic parameter to 1.0; put the training set into the object detection network and train for 200 epochs to obtain the optimal weights; Step (3): Detect the interference fringe pattern and draw the prediction box; input the test set divided in step (2) into the YOLOv7 network for detection to obtain the prediction box containing normalized position information; Step (4): Draw the center point on the interference fringe pattern; extract the prediction box containing the coordinates of the center point, obtain the original coordinates through inverse coordinate transformation and draw the center point of the interference fringe pattern, thereby realizing the high-precision positioning function.

2. The interferometric fringe pattern localization method based on target detection according to claim 1, characterized in that, The laser emits parallel polarized light with a wavelength of 532nm and a maximum output power of 3W.

3. The interferometric fringe pattern localization method based on target detection according to claim 1, characterized in that, A CCD camera with a resolution of 1920×1200 and a pixel size of 5.86μm was used as the receiving device.

Images