Method for correcting color distorted images of flue-cured tobacco leaves based on convolutional neural network

Abstract

This invention discloses a method for correcting color distortion images of roasted tobacco leaves based on a convolutional neural network. The method includes: acquiring a color-distorted image of roasted tobacco leaves; and inputting the color-distorted image into a convolutional neural network color correction model to obtain a corrected color image of the roasted tobacco leaves. This invention corrects the color distortion of tobacco leaves by inputting the color-distorted image into the convolutional neural network color correction model. According to the applicant's experiments, the image corrected using this invention can accurately identify the color of roasted tobacco leaves with an accuracy of 90%, and can be used as a basis for judging the roasting process of tobacco leaves, effectively solving the color distortion problem that occurs during the acquisition of tobacco leaf images.

Description

Technical Field

[0001] This invention relates to a method for correcting color distortion images of roasted tobacco leaves based on a convolutional neural network, belonging to the field of tobacco roasting technology. Background Technology

[0002] Currently, the determination of the degree of tobacco curing and the stage of the curing process mainly relies on the sensory perception of curing technicians to obtain the current color and shape of the tobacco leaves. By comparing these with standard samples, the curing process parameters are adjusted in a timely manner. However, tobacco curing processes based on human experience are not objective, have large individual differences, and result in unstable tobacco curing quality.

[0003] Based on the above, existing technology CN112540971A discloses a system and method for online acquisition of full information based on tobacco leaf characteristics. This system includes an RFID information acquisition module, an image acquisition module, an image processing module, a thickness measurement module, a weighing module, a near-infrared spectroscopy acquisition module, a chemical composition analysis module, a data classification module, and a data storage module. This solution can achieve online acquisition of full information on tobacco leaf characteristics, including image information, physical information, and chemical information, and can construct a full information database of tobacco leaf characteristics for convenient subsequent analysis and production use. However, the environment inside the tobacco curing chamber is complex and variable during the tobacco curing process, and the camera is in a high-temperature and high-humidity state, which causes color distortion during image acquisition, such as greenish images and blurry images, making accurate judgment difficult. Therefore, it is necessary to research and improve this approach. Summary of the Invention

[0004] Based on the above, the present invention provides a method for correcting color distortion images of roasted tobacco leaves based on convolutional neural networks, which can correct distorted images of roasted tobacco leaves in order to accurately determine the color of the tobacco leaves.

[0005] The technical solution of this invention is: a method for correcting color distortion images of roasted tobacco leaves based on convolutional neural networks, comprising:

[0006] Obtain a color-distorted image of roasted tobacco leaves;

[0007] The distorted color image of the tobacco leaves is input into a convolutional neural network color correction model to obtain a corrected color image of the roasted tobacco leaves.

[0008] Preferably, the method for constructing the convolutional neural network color correction model is as follows:

[0009] Acquire images of roasted tobacco leaves with normal and distorted colors;

[0010] The mean R, G, and B values ​​of non-zero pixels in the normal color image and the color-distorted image are extracted, and multimodal fusion is performed with the temperature and humidity information at the same time to establish an image color dataset;

[0011] Using the dataset of color-distorted images as input and the dataset of color-normal images as output, a multi-layer convolutional neural network is trained to obtain a convolutional neural network color correction model.

[0012] Preferably, before extracting the mean values ​​of R, G, and B, the normal color image and the color-distorted image of the tobacco leaf are segmented into a stem image and a leaf image.

[0013] Preferably, the convolutional neural network color correction model has three layers, wherein the first layer performs grayscale processing on the image, the second layer enhances the grayscale details of the image, and the third layer restores the color of the image through an end-to-end convolutional neural network. The output of the three layers is integrated to generate the final corrected image.

[0014] The beneficial effects of this invention are as follows: This invention corrects the color distortion of tobacco leaves by inputting the color distortion image of the tobacco leaf into a convolutional neural network color correction model. According to the applicant's test, the image corrected by this invention can accurately identify the color of tobacco leaf roasting with an accuracy of 90%, and can be used as a basis for judging the roasting of tobacco leaves, effectively solving the color distortion problem generated during the acquisition of tobacco leaf images. Attached Figure Description

[0015] Figure 1 This is a flowchart of a method for correcting color distortion images of roasted tobacco leaves based on a convolutional neural network. Detailed Implementation

[0016] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0017] See Figure 1 The present invention provides a method for correcting color distortion images of roasted tobacco leaves based on convolutional neural networks, comprising the following steps:

[0018] S1 acquires a color-distorted image of roasted tobacco leaves;

[0019] During the tobacco curing process, images of the tobacco leaves are captured. Images with color distortion, such as those that appear greenish or blurry, are used as images to be corrected. The tobacco leaf images should correspond to the temperature and humidity information at the time of the event.

[0020] S2 inputs the distorted color image of the tobacco leaves into the convolutional neural network color correction model to obtain the corrected color image of the roasted tobacco leaves.

[0021] The image to be corrected is input into a convolutional neural network color correction model for prediction and correction, thereby obtaining the corrected color image of the roasted tobacco leaves. Specifically, the construction method of the convolutional neural network color correction model is as follows:

[0022] S21, acquire normal color image and color-distorted image of roasted tobacco leaves;

[0023] We initially selected 10-20 nodes in the tobacco curing process and obtained images of normal and distorted tobacco colors during the curing process.

[0024] S22, extract the R, G, and B mean values ​​of non-zero pixels in the normal color image and the color-distorted image, and establish an image color dataset by performing multimodal fusion with the temperature and humidity information at the same time.

[0025] Thresholding segmentation was used to segment normal and distorted tobacco leaf images into stem and leaf images to obtain better target features and recognition results. The mean R, G, and B values ​​of non-zero pixels in the segmented normal and corresponding distorted images were extracted, and multimodal fusion was performed with simultaneous temperature and humidity information to establish an image color dataset.

[0026] S23, using the dataset of the color-distorted images as input and the dataset of the color-normal images as output, train a multi-layer convolutional neural network to obtain a convolutional neural network color correction model.

[0027] In this embodiment, the convolutional neural network color correction model has three layers. The first layer performs grayscale processing on the image, relying on a CNN to transform the image into an optimal grayscale channel image. The second layer enhances grayscale details of the image, relying on a CNN to remove noise and enhance image quality. The third layer restores the color of the image through an end-to-end convolutional neural network. The output of the three layers is integrated to generate the final corrected image. Through the aforementioned training, the parameters of each layer of the convolutional neural network color correction model can be obtained, and then the input image can be corrected. According to the applicant's experiments, the image corrected by this invention can accurately identify the color of tobacco leaf roasting with an accuracy of 90%, and can be used as a basis for judging tobacco leaf roasting, effectively solving the color distortion problem that occurs during the acquisition of tobacco leaf images.

[0028] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for correcting color distortion images of roasted tobacco leaves based on convolutional neural networks, characterized in that, include: Obtain a color-distorted image of roasted tobacco leaves; The distorted color image of the tobacco leaves is input into a convolutional neural network color correction model to obtain a corrected color image of the roasted tobacco leaves; The method for constructing the convolutional neural network color correction model is as follows: Acquire images of roasted tobacco leaves with normal and distorted colors; The mean R, G, and B values ​​of non-zero pixels in the normal color image and the color-distorted image are extracted, and multimodal fusion is performed with the temperature and humidity information at the same time to establish an image color dataset; The dataset of color-distorted images is used as input, and the dataset of color-normal images is used as output to train a multi-layer convolutional neural network to obtain a convolutional neural network color correction model. Before extracting the R, G, and B mean values, the normal color image and the color-distorted image of the tobacco leaf are segmented into stem image and leaf image. The convolutional neural network color correction model has three layers, where the first layer performs grayscale processing on the image, the second layer enhances the grayscale details of the image, and the third layer restores the color of the image through an end-to-end convolutional neural network. The output of the three layers is integrated to generate the final corrected image.

Images