CN-115908461-B - Paleobiological image generation method based on neural network

CN115908461BCN 115908461 BCN115908461 BCN 115908461BCN-115908461-B

Abstract

The invention discloses a neural network-based paleobiological image generation method which comprises the steps of obtaining a paleobiological image data set and a non-paleobiological image data set, preprocessing the paleobiological image data set and the non-paleobiological image data set, performing visual angle image segmentation processing on the preprocessed paleobiological image data set and the preprocessed non-paleobiological image data set, training according to segmentation results to generate a countermeasure network Style GAN, generating a first visual angle image by using the trained generated countermeasure network Style GAN, and splicing the first visual angle image and a second visual angle image obtained by segmentation of the visual angle images to generate a new paleobiological image. According to the invention, the paleobiological image data set and the non-paleobiological image data set are subjected to visual angle image segmentation and then trained to generate the countermeasure network Style GAN, so that a clear countermeasure image can be generated, and the image information is ensured to be more complete.

Inventors

RAO YUNBO
WANG YIWEN
Zeng Shaoning

Assignees

电子科技大学长三角研究院(湖州)

Dates

Publication Date: 20260505
Application Date: 20221114

Claims (4)

1. The paleobiological image generation method based on the neural network is characterized by comprising the following steps of: S1, acquiring a paleobiological image data set and a non-paleobiological image data set; S2, preprocessing a paleobiological image data set and a non-paleobiological image data set; S3, performing visual angle image segmentation processing on the preprocessed paleobiological image dataset and the non-paleobiological image dataset, and training according to segmentation results to generate a countermeasure network Style GAN; the step S3 specifically comprises the following sub-steps: S31, respectively carrying out segmentation processing on a left view image L, a middle view image C and a right view image R on the preprocessed paleobiological image dataset and the preprocessed non-paleobiological image dataset to obtain n groups of three view image datasets corresponding to the paleobiological image dataset and the non-paleobiological image dataset; S32, randomly extracting a group of three-view images from three-view image data sets corresponding to the paleobiological image data set and the non-paleobiological image data set, forming a pair of samples (L, R) by the left-view image L and the right-view image R, and taking the middle-view image C as a true value C of the pair of samples; s33, repeating the operation to process the remaining n-1 groups of three-view images in the three-view image dataset to obtain n pairs of samples (L, R) and n truth values C; s34, randomly dividing n pairs of samples (L, R) into a training sample set and a test sample set according to a proportion, training the constructed generated countermeasure network Style GAN by adopting the training sample set, optimizing parameters of the generated countermeasure network Style GAN, and obtaining the trained generated countermeasure network Style GAN; S4, generating a first visual angle image by using the trained generation countermeasure network Style GAN, and splicing the first visual angle image with a second visual angle image obtained by segmentation of the visual angle image in the step S3 to generate a new paleontological image; The step S4 specifically comprises the following steps: And sequentially inputting samples (L, R) in the test sample set into an optimal generator G to obtain a first visual angle image A, and splicing the first visual angle image A to the left and right sides of the true value C to generate a new paleobiological image.
2. The method for generating an paleobiological image based on a neural network according to claim 1, wherein the step S2 specifically comprises the following sub-steps: s21, performing Gaussian smoothing with a window size of 5 multiplied by 5 on the paleobiological image dataset and the non-paleobiological image dataset by adopting Frequency Tuned algorithm, and removing high-frequency information in the image data; S22, global contrast highlighting processing is carried out on the paleobiological image dataset and the non-paleobiological image dataset by adopting Luminance Contrast algorithm.
3. The neural network-based paleobiological image generating method according to claim 2, wherein the calculation formula of the significance of the pixels in the Frequency Tuned algorithm is as follows: Wherein, the Is the arithmetic mean of the pixels of the image, To perform gaussian blur on the original image, the term "distance" means the distance between two points.
4. The neural network-based paleobiological image generating method according to claim 2, wherein the calculation formula of the significance of the pixels in the Luminance Contrast algorithm is as follows: Wherein, the Is an image The pixel value of the middle pixel point k, Is a gray value.

Description

Paleobiological image generation method based on neural network Technical Field The invention relates to the technical field of paleobiological image processing, in particular to a paleobiological image generation method based on a neural network. Background The ancient organisms are living on the earth in ancient times, and exist in the geological ages of the earth history, most of the ancient organisms are extinct at present, only remains, remains and living remains of the ancient organisms are reserved by the nature under specific conditions, and the ancient organisms are reserved in fossil form through petrifaction to form ancient fossil. Ancient fossil is a non-renewable natural heritage as a trace of ancient preservation. Through the archaea dataset, humans can obtain information about the earth's ecology in the past over time. And the paleobiodata set can effectively interface paleobion and data with computer science and database programming. With the development of the internet and the progress of artificial intelligence, it is also effective to study the ancient biological markers using machine learning techniques. Databases of ancient biologies established today are largely divided into two categories. One type is a collection management type database, the contribution to science popularization is obviously larger than that to scientific research, and the data quality is difficult to guarantee. The other type is a research type database, and the research content mainly comprises biological paleogeographic research, paleoecological and morphological evolution analysis and the like. But none of these databases meets the requirements of computer vision research. Disclosure of Invention Aiming at the defects in the prior art, the invention provides an ancient biological image generation method based on a neural network. In order to achieve the aim of the invention, the invention adopts the following technical scheme: A method for generating an paleobiological image based on a neural network comprises the following steps: S1, acquiring a paleobiological image data set and a non-paleobiological image data set; S2, preprocessing a paleobiological image data set and a non-paleobiological image data set; S3, performing visual angle image segmentation processing on the preprocessed paleobiological image dataset and the non-paleobiological image dataset, and training according to segmentation results to generate a countermeasure network Style GAN; S4, generating a first visual angle image by using the trained generation countermeasure network Style GAN, and splicing the first visual angle image with a second visual angle image obtained by segmentation of the visual angle image in the step S3 to generate a new paleontological image. Optionally, step S2 specifically includes the following substeps: s21, performing Gaussian smoothing with a window size of 5 multiplied by 5 on the paleobiological image dataset and the non-paleobiological image dataset by adopting Frequency Tuned algorithm, and removing high-frequency information in the image data; S22, global contrast highlighting processing is carried out on the paleobiological image dataset and the non-paleobiological image dataset by adopting Luminance Contrast algorithm. Optionally, a calculation formula of the significance of the pixel in the Frequency Tuned algorithm is: Wherein I μ is the arithmetic mean of the image pixels, To perform gaussian blur on the original image, the term "distance" means the distance between two points. Optionally, a calculation formula of the significance of the pixel in the Luminance Contrast algorithm is: wherein I k is a pixel value of a pixel point k in the image I, and I i is a gray value. Optionally, step S3 specifically includes the following substeps: S31, respectively carrying out segmentation processing on a left view image L, a middle view image C and a right view image R on the preprocessed paleobiological image dataset and the preprocessed non-paleobiological image dataset to obtain n groups of three view image datasets corresponding to the paleobiological image dataset and the non-paleobiological image dataset; s32, randomly extracting a group of three-view images from three-view image data sets corresponding to the paleobiological image data set and the non-paleobiological image data set, forming a pair of samples (L, R) by the left-view image L and the right-view image R, and taking the middle-view image C as a true value C of the pair of samples; S33, repeating the operation to process the rest (n-1) groups of three-view images in the three-view image dataset to obtain n pairs of samples (L, R) and n truth values C; S34, dividing n pairs of samples (L, R) into a training sample set and a test sample set according to proportion randomly, training the constructed generated countermeasure network Style GAN by adopting the training sample set, optimizing parameters of the generated countermeasure network Style GAN