TWI701640B - Ovulation prediction method based on saliva crystallization - Google Patents

Abstract

An ovulation prediction method based on saliva crystallization, comprises: obtaining a pre-identified image for use as a basis for determining an ovulation state; performing a one-color conversion step on the pre-identified image, and obtaining a conversion by filtering a specific color the image, the converted image is beneficial to the pre-identification determination and the subsequent ovulation state identification; the converted image is subjected to a color detection step for confirming whether the pre-identified image is a valid image; and the converted image is subjected to an area selecting a step and obtaining an image block of interest; performing a crystallization state classification step, classifying the image of interest image by crystallization; and performing an ovulation state determination step, and locating the image block of interest obtained by the crystallization state classification step the image is judged by the final result, and the pre-identified image is judged to be an ovulation period or a non-ovulation period. The ovulation prediction method based on saliva crystallization of the present invention can ensure that the pre-identified image obtained by the user is an effective image, which can increase the recognition success rate and avoid misinterpretation.

Description

Ovulation prediction method based on saliva crystallization

The present invention relates to a method for predicting ovulation, especially a method for predicting ovulation based on saliva crystals.

At present, the most common way for women to check the status of ovulation is to go to the hospital, which can be known by professional doctors through urine tests or other more confidential testing methods. However, the aforementioned methods are not only inconvenient for personal reasons, but also often because of going to the hospital for examination. It takes a lot of time and complicated procedures. Therefore, with the development of image recognition technology, camera devices, and smart phones, there have been more and more convenient ways to perform self-detection through some simple devices, such as saliva detection to determine women The state of ovulation.

The conventional technology I569766 provides a saliva image recognition method for predicting the ovulation period of women, which can be used to automatically detect and analyze women's saliva images simply, quickly, safely and accurately, and automatically determine that the female is in the non-ovulatory period from the saliva images , One of the possible ovulation period and the ovulation period, thereby helping the user to predict the ovulation period.

However, the conventional technology I569766 uses the number density of black and white pixels and the threshold value to determine an image. The threshold value needs to be defined by the user or by analyzing the statistical data of the user after a period of time. Due to the crystalline pattern of dried saliva, there may be many dots or Fern-shaped, there may be very few dots or fern-shaped, so the recognition success rate with pixel density is not good, and the threshold value is not that every user understands to make adjustments by himself, but it is easy to cause recognition The result of the error, so it is necessary to improve.

One object of the present invention is to provide a method for predicting ovulation based on saliva crystals, which provides users with a self-determining ovulation status through saliva images.

Another object of the present invention is to provide a method for predicting ovulation based on saliva crystals, which provides a more accurate judgment of ovulation status.

Another object of the present invention is to provide a method for predicting ovulation based on saliva crystals, which allows users to obtain better saliva images for identification.

In order to achieve the above and other objectives, the method for predicting ovulation based on saliva crystals in one embodiment of the present invention includes: obtaining a pre-identified image, which is used as a basis for determining the ovulation state; and performing one color on the pre-identified image It is a conversion step, and a conversion image is obtained by filtering a specific color, and the conversion image is beneficial to the judgment of the pre-recognition qualified and subsequent ovulation status recognition; the conversion image is subjected to a color detection step to confirm the pre-recognition image Whether it is a valid image; perform a region selection step on the converted image and obtain an image block of interest; perform a crystallization state classification step to classify the crystallization state of the image block of interest; and perform an ovulation state determination step, The final result judgment is made on the image of the interest image block obtained by the crystallization state classification step, and it is judged whether the pre-identified image is an ovulation period or a non-ovulation period.

In some embodiments of the present invention, the color system conversion step further includes performing a step of reacquiring a pre-recognized image to re-acquire a pre-recognized image.

In some embodiments of the present invention, wherein the color system conversion step is additionally provided with a range threshold, and the range threshold is 0.6 to 0.9.

In some embodiments of the present invention, the region selection step adopts a binary search method to circle and select a circular image block of interest.

In some embodiments of the present invention, the crystallization state classification step is trained by using a ResNet (Residual Neural Network) learning model.

In some embodiments of the present invention, wherein the ovulation state determination step is additionally provided with a probability threshold, the probability threshold is 0.5 to 0.9.

In some embodiments of the present invention, after the region selection step and before the crystalline state classification step, another image segmentation step is performed to perform image segmentation on the image block of interest to obtain a number of segmented images. Improve recognition efficiency.

In some embodiments of the present invention, wherein the image division step is to divide the image block of interest into four sub-image blocks by a cross division method.

In some embodiments of the present invention, wherein the ovulation state determination step is additionally provided with a number threshold.

In some embodiments of the present invention, the number threshold is 1~3.

In some embodiments of the present invention, after the region selection step and before the crystalline state classification step, another image compression step is performed to compress the image block of interest to improve the recognition efficiency.

In some embodiments of the present invention, the crystallization state classification step further includes performing a data amplification step to obtain more image data with the same characteristics to improve the recognition accuracy.

S1: Image capture step

S2: Color system conversion steps

S3: Color detection step

S4: Region selection steps

S41: Image segmentation step

S42: Image compression step

S5: Steps to classify crystalline state

S50: Data amplification step

S6: Steps to identify ovulation status

ρ : range threshold

θ : probability threshold

λ : number threshold

μ : crystallization probability

γ : Prediction probability of ovulation period

ε: error value

R: Effectiveness threshold

Fig. 1 is a flowchart of an embodiment of a method for predicting ovulation based on saliva crystallization; Fig. 2 is a flowchart of another embodiment of a method for predicting ovulation based on saliva crystallization; Fig. 3 is a kind of one of the present invention A flowchart of another embodiment of a method for predicting ovulation based on saliva crystallization; Figure 4 is a flowchart of another embodiment of a method for predicting ovulation based on saliva crystallization; Figure 5 is a flowchart of another embodiment of ovulation prediction based on saliva crystallization A flowchart of another embodiment of the method; Figure 6-1 is a non-saliva crystallization image of a saliva crystallization-based ovulation prediction method of the present invention; Figure 6-2 is one of the saliva crystallization-based ovulation prediction methods of the present invention The HSV black conversion diagram of the non-saliva crystal image; Figure 6-3 is the white conversion diagram of the non-saliva crystal image HSV which is one of the saliva crystal-based ovulation prediction methods of the present invention; Fig. 6-4 is a saliva crystal-based ovulation of the present invention One of the prediction methods is a non-saliva crystallization image HSV red conversion diagram; Figure 6-5 is a non-saliva crystallization image HSV green conversion diagram of a saliva crystallization-based ovulation prediction method of the present invention; Figure 6-6 is a kind of based on the present invention One of the saliva crystallization prediction methods for ovulation, the HSV blue transition image of non-saliva crystallization images; Fig. 7-1 is a saliva crystal image, one of the saliva crystal-based ovulation prediction methods of the present invention; Fig. 7-2 is a saliva crystal image HSV black conversion diagram, one of the saliva crystal-based ovulation prediction methods of the present invention; Fig. 7 -3 is a saliva crystal image HSV white conversion diagram of one of the saliva crystal-based ovulation prediction methods of the present invention; Fig. 7-4 is a saliva crystal image HSV red conversion diagram of one of the saliva crystal-based ovulation prediction methods of the present invention; Fig. 7-5 is a green conversion diagram of saliva crystal image HSV, one of the saliva crystal-based ovulation prediction methods of the present invention; Fig. 7-6 is a saliva crystal image HSV blue color, one of the saliva crystal-based ovulation prediction methods of the present invention Conversion diagram; Figure 8 is a saliva crystal-based ovulation prediction method without fern-shaped saliva crystal images; Figure 9 is a saliva crystal-based ovulation prediction method with fern-shaped saliva crystal images Fig. 10 is a flow chart of ovulation status identification of a saliva crystal-based ovulation prediction method of the present invention; Fig. 11 is a data training process and a saliva crystal classifier generation diagram of a data training process of a saliva crystal-based ovulation prediction method of the present invention.

FIG. 1 is a flowchart of an embodiment of a method for predicting ovulation based on saliva crystallization according to the present invention. Please refer to FIG. 1. An ovulation prediction method based on saliva crystals of the present invention includes: an image capture Take step S1, a color system conversion step S2, a color detection step S3, a region selection step S4, an image segmentation step S5, a crystal classification step S6, and an ovulation state identification step S7. The image capturing step S1 is to obtain a pre-identified image. The image capturing step S1 can be implemented by any camera device, such as a camera on a smart phone plus a microscope lens, after applying saliva on a glass slide The pre-identified image is obtained by the user pointing the lens at the glass slide to shoot.

The color system conversion step S2 converts the pre-recognized image into a converted image to reduce recognition errors and facilitate subsequent image recognition procedures. The conversion of the converted image facilitates subsequent image digitization processing. The color system conversion system uses a common method in color science, namely, the conversion between the RGB color model (the three primary colors of Red, Green, and Blue) and the HSV color space model (Hue, Saturation, Value).

Generally, an image represented by an RGB color model uses three values to describe the same pixel. In a 24-bit image, each element is represented by a number between 0 and 255. The HSV (Hue, Saturation, Value) color space model corresponds to a cone-shaped subset in the cylindrical coordinate system. The top surface of the cone corresponds to V=1, which contains R=1 and G=1 in the RGB color model. , B=1 three faces, the colors represented are brighter. Generally, when converting RGB image to HSV color space, the value range of H channel is between 0~180, the value range of S channel is between 0~255, and the value range of V channel is between 0 Between ~255, the conversion formula from RGB to HSV is as follows:

v = max

In the above formula, h = the value of the H channel, s = the value of the S channel, v = the value of the V channel, r = the value of red in RGB, g = the value of green in RGB, b = the value of blue in RGB, Table 1 shows the range of common HSV basic color components. The above-mentioned conversion method is the basic technique of general image processing, and will not be repeated here.

In this embodiment, the HSV color component upper limit (max) and lower limit (min) ranges used in the color system conversion step S2 are shown in Table 2 below:

In this embodiment, the color system conversion step S2 performs five color conversions on the pre-identified image through HSV color conversion, which are five colors of black, white, red, green, and blue, to obtain a converted image. The image only contains the above five color components of black, white, red, green and blue. The converted image is beneficial to the judgment of the pre-recognition image and the subsequent ovulation status identification. Of course, other colors and color numbers can be performed according to needs. Conversion, the present invention is not limited.

Then proceed to a color detection step S3, the color detection step S3 is to detect by setting Black and green. In this embodiment, it is used to determine whether the black or green distribution coverage in the converted image is the largest or the second largest to perform image screening to ensure that the pre-identified image is a qualified image. It is indeed an image of saliva. In this embodiment, it is first judged whether the color of the largest coverage area in the converted image is black or green. If not, it means that the pre-identified image is a non-qualified image, that is, a non-saliva image, and a new image must be taken again. , That is, re-execute the image capturing step S1, if yes, continue to determine whether the color of the second largest coverage area is one of black or green, if not, it means that the pre-identified image is a non-qualified image and must be re-shot to obtain a new Image, that is, re-execute the image capturing step S1, if yes, proceed to a region selection step S4. The selection of the detected color in the aforementioned converted image can be determined according to actual requirements, and the present invention is not limited.

In image processing technology, the specific area of the image is called Regions Of Interest or ROIs, and this aspect of processing is called ROI processing. In this embodiment, when the region selection step S4 is performed, an ROIs processing module is used to perform image edge detection on the converted image by a binary search method to obtain a saliva crystallized region in the converted image and draw a circle The saliva crystallized area is circled to form a circular image block. The specific image area circled in step S4 through the area selection step S4 will facilitate subsequent identification, that is, only identify the specific area of interest, that is, Only the circular image block needs to be identified, and there is no need to identify the area outside the circular image block, because the ovulation state is judged by the shape of the saliva crystals. There are many edge detection methods, such as Otsu algorithm, Sobel edge detection, and Hough circle detection. In this example, a binary search method is used, but the invention is not limited.

After the region selection step S4 is completed, a crystallization state classification step S5 is performed to classify the crystalline state of the circular image block of interest obtained in the foregoing steps. The classification method used here is based on a neural network model for learning Training, such as: VGG, Inception, ResNet (Residual Neural Network)... and other models, this is a well-known neural network model, and will not be detailed here.

In this embodiment, the ResNet deep learning model is used as the image recognition training model of the present invention. In the crystallization state classification step S5, a saliva crystal classifier obtained from the image recognition training model is used in advance, and the saliva crystal classifier is used to classify images of ovulatory crystal and non-ovulatory crystal state, and the saliva crystal classifier will The circular interest image is classified into ovulation crystals and non-ovulatory crystals to facilitate the subsequent determination of ovulation and non-ovulation phases. If it is classified as non-ovulatory crystals, the classification of the interest image block is set to "0", If it is classified as an ovulation crystal, the classification of the image block of interest is set to "1", and then an ovulation state determination step S6 is performed.

After the crystallization state classification step S5 is performed, the classification result of the interesting image block is passed through the ovulation state judgment step S6, and the ovulation state judgment result is displayed to inform the user whether it is an ovulation state or a non-ovulation state for User reference. Through the above method, a more effective method for judging the ovulation period can be provided.

Please continue to refer to Figure 1. Preferably, a range threshold ρ is additionally provided in the color detection step S3, and the range threshold ρ is preferably between 0.6 and 0.9 to improve the recognition accuracy of qualified images. In this embodiment, the range threshold ρ = 0.7. When performing the color detection step S3, calculate the coverage rates of the five colors of black, white, red, green and blue in the entire image respectively, and calculate the largest and second largest color coverage rates Whether it is black or green, if not, re-execute the image capturing step S1, if yes, then add the coverage of black and green, and calculate the sum of the coverage of both and the sum of the coverage of all colors Whether the ratio is greater than or equal to 0.7, if not, re-execute the image capturing step S1, if yes, then proceed to the region selection step S4. Wherein, the range threshold ρ described in this embodiment can be set according to actual requirements, and the present invention is not limited.

Please continue to refer to Figure 1. In the crystallization state classification step S5, the saliva crystallization image is primarily classified through the neural network-like learning model. In the ovulation state determination step S6, in order to increase the accuracy of the ovulation state determination, another set there is a probability threshold value [theta], is performed based on the final ovulation state determination through the probability threshold value [theta], wherein the probability threshold value [theta] is preferably 0.5 to 0.9, in this embodiment the probability threshold value [theta] according to the present embodiment is 0.8, i.e., when the circular interest When the probability of the ovulation state of the image is determined to be greater than or equal to 0.8 after learning by a neural network training model, it is finally determined that the pre-identified image belongs to the ovulation period, otherwise, it is not, that is, it is finally judged that the pre-identified image belongs to the non-ovulatory period.

2 is a flowchart of another embodiment of a method for predicting ovulation based on saliva crystals according to the present invention, please refer to FIG. 2. Preferably, after the region selection step S4 and before the crystalline state classification step S5 is performed, another image segmentation step S41 is performed to perform image segmentation on the image block of interest to obtain multiple segmented images, which can be enlarged after segmentation The image feature data can improve the recognition efficiency and recognition accuracy. In this embodiment, the image block of interest is divided into four sub-image blocks by the cross division method, which are a first sub-image block and a second sub-image block. Two sub-image blocks, a third sub-image block, and a fourth sub-image block. Each sub-image block has a size of 500×500. The size of the sub-image block can be set according to actual needs. The invention is not limited.

Please continue to refer to Figure 2. Preferably, another number threshold λ is set in the ovulation state determination step S6. In this embodiment, since the image block of interest is divided into four sub-image blocks, the number threshold λ is set to 1~ 3, wherein the threshold number is an integer of [lambda], can be based on needs (such as the number of division images) sets the threshold number of [lambda], so when performing the classification step S5 classified crystalline state, the first sub-image block, the second sub-image The block, the third sub-image block, and the fourth sub-image block will be sequentially classified into the crystalline state, each being classified as "1" or "0", and then the ovulation state determination step S6 is executed , The probability threshold θ and the quantity threshold λ can be used for more efficient and accurate ovulation status determination.

FIG. 10 is a flowchart of ovulation status identification of a saliva crystal-based ovulation prediction method of the present invention. Please also refer to FIG. 10. As described above, in one embodiment, when the ovulation state determination step S6 is performed, when the probability threshold θ = 0.8 and the number threshold λ is set to 1, that is, there is one sub image in the four sub image blocks When the crystallization state of the block is classified as "1" and its crystallization probability μ is greater than or equal to 0.8, it is judged that the pre-identified image belongs to the ovulation period; when the probability threshold θ = 0.7 and the number threshold λ is set to 2, that is When the crystallization state of two of the four sub-image blocks is classified as "1", although the crystallization probability μ is less than 0.8 but both are greater than or equal to 0.7, it is also judged that the pre-identified image belongs to the ovulation period; Threshold θ = 0.6, when the number threshold λ is set to 3, that is, when the crystallization state of three of the four sub-image blocks is classified as "1", although the crystallization probability μ is less than 0.7 but all are greater than or equal to 0.6, it is also judged that the pre-identified image belongs to the ovulation period, where the crystallization probability μ in each of the above situations is the probability calculated when the crystallization state of a sub-image block is classified as "1". Conversely, if the above verification method is not met, it is determined that the pre-identified image belongs to the non-ovulatory period. Wherein, the probability threshold θ and the quantity threshold λ described in this embodiment can be set according to actual requirements, and the present invention is not limited.

其中，μ _k

{μ|μ

θ}，

μ _k ：0

μ _k

1，n=|{μ}|，1

n

4，0.001

ε

0.025，其中ε為一誤差值，該誤差值ε介於一誤差值範圍之間，可以實際需求調整設定誤差值範圍並給定該誤差值ε一設定值。 As mentioned above, in this embodiment, after the four sub-images are all crystalline and classified, if it is judged to belong to the ovulation state, an ovulation prediction probability γ is calculated for the user's reference, and the ovulation period The prediction probability γ is calculated as follows:

Among them, μ _k

{ μ | μ

θ },

μ _k :0

μ _k

1, n =|{ μ }|, 1

n

4, 0.001

ε

0.025, where ε is an error value, and the error value ε is between an error value range. The error value range can be adjusted according to actual requirements and the error value ε is given a set value.

，即表示排卵期的機率為80.25%。 For example, when the probability threshold θ = 0.8, the number threshold λ is set to 1, the error value ε = 0.001, assuming that the crystallization probability of one of the sub image blocks is μ = 0.85 and n =1, that is, μ ₁ = 0.85 ,then

, Which means that the probability of ovulation is 80.25%.

，即表示排卵期的機率為71% For another example, when the probability threshold θ = 0.7, the number threshold λ is set to 2, the error value ε = 0.001, assuming that the crystallization probability μ of two sub-image blocks are 0.72 and 0.76 respectively, and n = 2, That is, μ ₁ =0.72, μ ₂ =0.76, then

, Which means the probability of ovulation is 71%

FIG. 3 is a flowchart of another embodiment of a method for predicting ovulation based on saliva crystallization according to the present invention. Please refer to FIG. 3. Preferably, after the region selection step S4 and before the crystalline state classification step S5 is performed, another image compression step S42 is performed to perform image compression on the image block of interest to improve the recognition efficiency. Because when the deep learning model makes predictions, the image is usually compressed to the size with the best recognition effect of the model. For example, the best image size of GoogLeNet is 224×224, the best image size of Inception V3 is 299×299, and the best image size of YOLO v3 The best image size is 416×416. This embodiment uses the best size 224×224. Therefore, before performing the crystalline state classification step S5, perform the image compression step S42 to provide the best size of the input image of the crystalline classifier. It will be helpful for subsequent image classification. Of course, the image compression step S42 can also be executed after the image segmentation step S41 is executed to increase the classification and overall identification efficiency. Among them, the sub-image compression size can be set according to actual needs. The present invention does not limit.

4 is a flowchart of another embodiment of a method for predicting ovulation based on saliva crystals according to the present invention, please refer to FIG. 4. Preferably, when the image segmentation step S41 is executed and then the image compression step S42 is executed, a better image result will be obtained, which is more conducive to the efficiency in the recognition process.

FIG. 5 is a flowchart of another embodiment of a method for predicting ovulation based on saliva crystallization according to the present invention. Please refer to FIG. 5. Preferably, before entering the neural network model training, data amplification can be performed on the image database, that is, more image data can be input in advance to make it easier for the neural network to find distinguishable features, mainly because it can avoid data Too little, resulting in situations where it is impossible to distinguish important features, resulting in Misjudgment, that is, if the image features cannot be clearly distinguished, it is easy to judge different but similar conditions as the same condition. Through the data amplification step S50, the crystallization state classification step S5 can be executed to make the interest image perform ovulation and non-ovulation. The preliminary classification of ovulation status is more accurate.

Figures 6-1 to 6-6 are the HSV color conversion diagrams of non-saliva crystallization images, one of the saliva crystallization-based ovulation prediction methods of the present invention, please refer to Figures 6-1 to 6-6. In an embodiment of the present invention, Figure 6-1 is a landscape image. After HSV color system conversion, the black conversion map of Figure 6-2 is obtained, and the black distribution coverage rate is 16532.5, Figure 6 3 White conversion map, its white distribution coverage rate is 42088.5, Figure 6-4 red conversion map, its red distribution coverage rate is 57323.5, Figure 6-5 green conversion map, its green distribution coverage rate is 174848.5, Figure 6-6 blue The color conversion map has a blue coverage rate of 3739.5. It can be seen from Figures 6-2 to 6-6 that the color of the largest coverage area is green, and the color of the second largest coverage area is red. Therefore, when the color detection step S3 is performed, the pre-identified image is determined If it is a non-qualified image, a new image must be captured again, that is, the image capturing step S1 is executed again.

Figures 7-1~Figure 7-6 are the HSV color conversion diagrams of saliva crystal images, one of the saliva crystal-based ovulation prediction methods of the present invention, please refer to Figure 7-1~Figure 7-6. In an embodiment of the present invention, Figure 7-1 is a saliva crystal image. After HSV color system conversion, the black conversion map of Figure 7-2 is obtained, and the black distribution coverage rate is 5788509.5, Figure 7 -3 White conversion map, its white distribution coverage rate is 0, Figure 7-4 red conversion map, its red distribution coverage rate is 4293821.0, Figure 7-5 green conversion map, its green distribution coverage rate is 10322792.5, Figure 7-6 The blue conversion graph has a blue distribution coverage rate of 0. From Figure 7-2~Figure 7-6, the color of the largest coverage area is green, and its distribution coverage rate is 10322792.5, and the color of the second largest coverage area is black, and its distribution coverage rate is 5788509.5. Therefore, in the implementation In the color detection step S3, it is determined that the pre-identified image is a qualified image, and then the region selection step S4 is executed to obtain an image block of interest.

0.78957，0.78957大於該圍閾值ρ，因此判斷該預辨識影像為合格之影像，接著執行該區域選取步驟S4，當然該範圍閾值ρ可依照當下需求調整。 Please continue to refer to Figure 7-1~Figure 7-6. In some of the foregoing embodiments, preferably, a range threshold ρ is set when performing the color detection step S3 to ensure that the judgment of qualified images can be more accurate. Set the range threshold ρ = 0.7, and the converted image is black The ratio of the distribution coverage rate and the green distribution coverage rate to the total color is (5788509.5+10322792.5)/20405123.0

0.78957, 0.78957 is greater than the surrounding threshold ρ, so it is determined that the pre-identified image is a qualified image, and then the region selection step S4 is performed. Of course, the range threshold ρ can be adjusted according to current needs.

FIG. 8 is an image of saliva crystals without fern-shaped saliva in a saliva crystal-based ovulation prediction method of the present invention. Please refer to FIG. 8. Under normal circumstances, saliva in the ovulation period will have a fern-like crystal shape, but not in the non-ovulation period. Figure 8(a) is a converted image of the region selection step S4 that has not yet been performed, and Figure 8(b) is an implementation of this The converted image in the area selection step S4, that is, the circled area in FIG. 8(b) is an image block of interest.

FIG. 9 is an image of a fern-shaped saliva crystal with a saliva crystal-based ovulation prediction method of the present invention. Please refer to FIG. 9. Among them, Figure 9(a) is a converted image that has not yet performed the region selection step S4, and Figure 9(b) is a converted image that has performed the region selection step S4, that is, the circled area in Figure 9(b) is an interest Image block.

FIG. 11 is a data training process and a diagram of a saliva crystal classifier generated by a saliva crystal-based ovulation prediction method of the present invention, please refer to FIG. 11. When performing machine learning model training, first provide a training image data, and the training image data has a certain amount according to the training requirements, the training image data is the so-called experimental group image, and then these images are processed by ROIs to find the image range of interest , And then perform image segmentation. The image is divided into several sub-images as required. Image segmentation can amplify the feature information, thereby increasing the efficiency and accuracy of subsequent training, and then classify ovulation crystals and non-ovulation crystals. Here the so-called ovulation crystals The image of the non-ovulatory crystal is confirmed in advance, and then the image compression is performed Then it is thrown into the neural network model for machine learning, and finally a crystalline classifier is produced. The crystalline classifier can be used to determine whether it is ovulation based on the state of saliva crystallization.

Continue to refer to FIG. 11, when the crystal classifier is obtained, the performance evaluation test of the crystal classifier is performed through a test image data, and the test image data has a certain amount. The test image data is the so-called control image. The data is an image data set that has undergone ROIs image processing (interest image processing), image segmentation, ovulation and non-ovulation classification and compression. When performing performance evaluation, a performance threshold R is set, and the performance threshold R is preferably 0.6~ 0.9, when the performance evaluation test result is greater than or equal to the performance threshold R, it means that the ovulation status recognition effect of the crystal classifier meets expectations, otherwise it is not. For example, when the performance threshold R is equal to 0.6, it means that the crystal classifier is expected to be The ovulation status recognition success rate is more than 60%. Therefore, if the performance evaluation test result of the crystal classifier and the test image data is 0.7, it means that the ovulation status recognition success rate of the crystal classifier is 70%, which is in line with expectations. The structure and parameters of this type of neural network model are stored for use in subsequent saliva crystal identification. On the contrary, if the performance evaluation test result of the crystal classifier and the test image data is 0.5, it means the ovulation state of the crystal classifier If the recognition success rate is 50%, if it does not meet expectations, it will return to the neural network model for machine learning and adjust the learning parameters to retrain.

The above-mentioned embodiments are only to illustrate the technical ideas and features of the present invention, and their purpose is to enable those who are familiar with the art to understand the content of the present invention and implement them accordingly. However, the patent scope of the present invention cannot be limited by this. All the equal changes or modifications made in accordance with the spirit of the present invention and the contents of the specification shall be covered by the scope of the patent of the present invention.

S1: Image capture step

S2: Color system conversion steps

S3: Color detection step

S4: Region selection steps

S5: Steps to classify crystalline state

S6: Steps to identify ovulation status

Claims (11)

A method for predicting ovulation based on saliva crystallization, including: obtaining a pre-identified image; performing a color conversion step on the pre-identified image, performing color conversion on the pre-identified image through HSV color conversion, and obtaining a converted image after filtering a specific color ; Perform a color detection step on the converted image to determine whether the black or green distribution coverage in the converted image is the largest or second largest for image screening to confirm whether the pre-identified image is a valid image; Perform a region selection on the converted image, perform image edge detection on the converted image, obtain a saliva crystallized area in the converted image to obtain an image block of interest; perform a crystallization state classification step to perform the image block of interest Carrying out crystallization state classification; and performing an ovulation state determination step, the final result determination of the image of the interest image block obtained by the crystallization state classification step, and determining whether the pre-identified image is an ovulation or non-ovulation period. The method for predicting ovulation based on saliva crystals as described in claim 1, wherein the step of performing the color system conversion further includes performing a step of reacquiring a pre-identified image to re-acquire a pre-identified image. The method for predicting ovulation based on saliva crystals as described in claim 1, wherein the color detection step is additionally provided with a range threshold, the range threshold is 0.6 to 0.9, and the color detection step includes the conversion Add the coverage of black and green in the image, and calculate whether the ratio of the sum of the coverage of the two to the sum of the total color coverage is greater than or equal to the range threshold. The method for ovulation prediction based on saliva crystals described in the first item of the patent application, wherein the region selection step adopts a binary search method to circle and select a circular image block of interest. According to the ovulation prediction method based on saliva crystals as described in item 1 of the scope of patent application, the crystallization state classification step is trained by using a ResNet (Residual Neural Network) learning model. The method for predicting ovulation based on saliva crystals as described in item 1 of the scope of patent application, wherein the ovulation state determination step is additionally provided with a probability threshold, and the probability threshold is 0.5 to 0.9. The method for predicting ovulation based on saliva crystals as described in claim 1, wherein after the region selection step and before performing the crystallization state classification step, another image segmentation step is performed to perform the image block of interest Image segmentation, to obtain multiple segmented images to improve recognition efficiency. The method for predicting ovulation based on saliva crystals as described in item 7 of the scope of patent application, wherein the image segmentation step is to divide the image block of interest into four sub-image blocks by a cross segmentation method. The method for predicting ovulation based on saliva crystals as described in item 6 of the scope of patent application, wherein the ovulation state determination step is additionally provided with a number threshold, and the number threshold is 1~3. The method for predicting ovulation based on saliva crystals as described in claim 1, wherein after the region selection step and before the crystallization state classification step, another image compression step is performed to perform the image block of interest Image compression to improve recognition efficiency. The method for ovulation prediction based on saliva crystals as described in claim 1, wherein the crystallization state classification step further includes performing a data amplification step to obtain more image data with the same characteristics to improve the recognition accuracy rate.

Images