Hyperspectral signal analysis of blended materials

Abstract

This invention provides a hyperspectral analysis method for blended materials, comprising the following steps: capturing first information of the test object using an image capturing unit and transmitting it to a control unit, the first information including optical image and spectral reflectance; preprocessing the first information to generate second information; obtaining multiple feature values ​​of the second information using a convolutional neural network; judging and generating a judgment result using a deep learning model and the feature values; and classifying the test object according to the judgment result to determine the ratio of polyester fiber to pure cotton in the test object.

Description

Technical Field

[0001] This invention relates to a hyperspectral analysis method for blended materials, and more particularly to a method and system for adjusting the drive wheels of a vehicle through steering input and throttle valve input. Prior Technology

[0002] Hyperspectral signals refer to signals composed of the collection of reflected energy sensed in the wavelength range of 300 nanometers to 2500 nanometers. Hyperspectral images are image data obtained by acquiring hyperspectral signals using a two-dimensional array. Such images contain hyperspectral signals for each pixel, thus each pixel possesses spectral information from many different wavelengths. Different substances have unique spectral characteristics, which can be utilized to develop various hyperspectral technologies.

[0003] Hyperspectral technology has shown significant application potential in various industries and provides more comprehensive and efficient solutions for production and management.

[0004] In agriculture, hyperspectral technology can monitor plant growth, detect diseases and nutritional status, helping farmers optimize agricultural production. In environmental monitoring, hyperspectral imaging technology can identify pollutants in air and water, providing real-time monitoring and early warning, which helps maintain ecological balance and protect the environment and human health.

[0005] In textile production, hyperspectral technology can be used to detect the quality of textile raw materials, including textile materials and dyes. Through hyperspectral imaging, the composition and quality of textiles can be analyzed quickly and non-destructively, which helps to improve product consistency and quality.

[0006] In the medical field, hyperspectral technology can be used for biomedical imaging, disease diagnosis, and monitoring of treatment effects, providing doctors with more accurate diagnostic tools.

[0007] Hyperspectral technology can capture the spectral characteristics of objects' reflection or emission, possessing up to 200 continuous spectral bands, effectively improving spectral resolution. This enables hyperspectral sensors to detect, classify, differentiate, identify, and quantify minute objects and substances in a non-invasive manner. However, when many such targets may be difficult to identify beforehand or visually inspect, we can use the spectral characteristics in hyperspectral images for quantitative analysis, including estimating the concentration ratio of substances using the intensity or location of spectral peaks and the shape of the spectrum. Through quantitative analysis of spectral characteristics, we can understand the chemical composition, surface properties, and other information of the target object, thereby achieving accurate classification and identification of the target object.

[0008] The textile industry is an important industry in Taiwan. Due to its diversity and complexity, traditional quality assessment and analysis are difficult. The quality and safety of products are usually defined by physical characteristics (such as texture, color, pattern, etc.) and chemical characteristics (such as chemical composition, solubility, reaction rate, pH value, leakage rate).

[0009] Traditional quality and safety assessments require manual visual inspection and cumbersome, time-consuming, destructive, and environmentally harmful chemical or biological tests. Therefore, there is an urgent need for accurate, rapid, real-time, and non-chemical testing technologies to optimize quality and ensure product safety. Hyperspectral imaging technology has become a widely used method in quality and safety assessments in manufacturing.

[0010] While hyperspectral technology offers an effective way to monitor and evaluate textile materials, the presence of hyperspectral images and the mixing of multiple materials complicates unmixing analysis. Therefore, it is crucial to develop a hyperspectral unmixing quantitative analysis module specifically designed for mixed textile materials.

[0011] Therefore, a hyperspectral analysis method for blended materials is provided, which can identify blended materials by capturing optical images and spectra of the test object, combining feature extraction of convolutional neural networks and prediction of deep learning, and is a problem that those skilled in the art wish to solve. Summary of the Invention

[0012] One objective of this invention is to provide a hyperspectral analysis method for blended materials. This method involves capturing optical images and spectra of the test material, extracting features from the preprocessed optical images and spectra using a convolutional neural network, and then making a judgment based on a deep learning model to achieve the effect of identifying blended materials.

[0013] To achieve the above objectives, the present invention provides a hyperspectral analysis method for blended materials, comprising: acquiring first information of a test object using an image acquisition unit, the first information including an optical image and a spectral reflectance; transmitting the first information to a control unit; preprocessing the first information to generate second information; obtaining a plurality of feature values ​​of the second information using a convolutional neural network; judging the feature values ​​based on a deep learning model to generate a judgment result; and classifying the test object based on the judgment result to determine the proportion of polyester fiber and pure cotton constituting the test object.

[0014] The present invention provides an embodiment in which the step of the control unit preprocessing the first information to generate a second information further includes the steps of: the control unit segmenting the first information to generate a plurality of segmented information; and the control unit classifying the segmented information through a spectral angle matching algorithm to generate the second information.

[0015] The present invention provides an embodiment in which, in the step of using an image capturing unit to capture first information of a test object, the first information including an optical image and a spectral reflectance, wherein one wavelength range of the spectral reflectance is 900 nanometers to 1700 nanometers.

[0016] The present invention provides an embodiment in which, in the step of the control unit judging based on a deep learning model and the feature values ​​to generate a judgment result, the judgment result corresponds to the ratio of polyester fiber to pure cotton in one of the test objects.

[0017] The present invention provides an embodiment in which, in the step of the control unit classifying the test object according to the judgment result, the classification is divided into a plurality of categories in which the proportion of polyester fiber and pure cotton is 0% / 100%, 20% / 80%, 35% / 65%, 50% / 50%, 65% / 35%, 80% / 20%, and 100% / 0%.

[0018] Therefore, a hyperspectral analysis method is provided that can automatically identify analytes with different compositions without damaging the analytes in blended materials, which is a problem that those skilled in the art wish to solve. Simple Explanation of the Diagram

[0019] Figure 1: It is a flowchart of the hyperspectral analysis method for the blended materials of the present invention; Figure 2: It is a schematic diagram of the hyperspectral analysis system for the blended material of the present invention; Figure 3: It is a schematic diagram of the steps of the hyperspectral analysis method for the blended material of the present invention; Figure 4: It is a schematic diagram of the steps of the hyperspectral analysis method for the blended material of the present invention; Figure 5: It is a schematic diagram of the steps of the hyperspectral analysis method for the blended material of the present invention; Figure 6: This is a flowchart of the hyperspectral analysis method for the blended materials of the present invention; and Figure 7: It is a schematic diagram of the steps of the hyperspectral analysis method for the blended material of the present invention. Implementation

[0020] To enable your review committee to gain a better understanding of the features and effects achieved by this invention, preferred embodiments and detailed descriptions are provided below:

[0021] Currently, the testing of textile materials produced by mixing multiple materials mostly uses destructive single-point testing methods. This requires destroying a portion of the textile material and testing the destroyed point. Furthermore, due to the large production volume of textile materials, it is impossible to test all textile materials one by one. Only sampling inspection can be carried out. As a result, the inspection results can only be confirmed as estimated values, and it is impossible to confirm the material composition of all textile materials.

[0022] This invention combines convolutional neural networks and deep learning models to capture the light reflected by the test object, generating optical images and spectra. It then investigates the spectral reflectance and feeds this data into the convolutional neural network for feature extraction. The deep learning model then evaluates the feature values ​​to detect the composition of the textile material. This allows for the detection of the composition of the test object without damaging it. Furthermore, since only the optical image and spectrum of the test object are needed, a single test object does not require excessive time. This enables the detection of all test objects, ensuring both qualitative and quantitative quality indicators.

[0023] The invention will be described in detail below by way of the drawings illustrating various embodiments thereof. However, the concept of the invention may be embodied in many different forms and should not be construed as being limited to the exemplary embodiments set forth herein.

[0024] First, please refer to Figure 1, which is a flowchart of the hyperspectral analysis method for blended materials according to the present invention. As shown in the figure, the steps of the hyperspectral analysis method for blended materials according to the present invention will be described in detail below:

[0025] Step S10: Use the image capturing unit to capture the first information of the object under test, the first information including optical image and spectral reflectance;

[0026] Step S20: The image capturing unit transmits the first information to the control unit;

[0027] Step S30: The control unit preprocesses the first information to generate the second information;

[0028] Step S40: The control unit uses a convolutional neural network to obtain multiple feature values ​​of the second information;

[0029] Step S50: The control unit determines these feature values ​​based on the deep learning model to generate a judgment result;

[0030] Step S60: The control unit classifies the test object according to the judgment result;

[0031] Next, please refer to Figure 2, which is a schematic diagram of the hyperspectral analysis system for blended materials of the present invention. As shown in the figure, a hyperspectral analysis system 1 for blended materials of the present invention will be described in detail. It includes: a test object 10; an image capturing unit, which captures an optical image and a spectral reflectance of the test object; a light source 30, which illuminates the test object 10; and a control unit 40, which is connected to the image capturing unit 20 and the light source 30. The control unit 40 includes: a convolutional neural network 42 and a deep learning model 44.

[0032] The following will describe each step in detail. Please refer to Figures 2 and 3. Figure 3 is a schematic diagram of the steps of the hyperspectral analysis method for blended materials of the present invention. As shown in the figure, in step S10, an image capturing unit 20 captures first information 50 of a test object 10. The first information 50 includes an optical image 52 and a spectral reflectance 54. In this embodiment, the light source 30 includes an adjustable working mechanism (not shown), which is used to adjust the light source 30 to adapt to the test object 10 of various shapes. The light source 30 is adjusted by the adjustable working mechanism so that the test object 10 can receive sufficient light, so that the optical image 52 and the spectral reflectance 54 in the first information 50 can be appropriately captured by the image capturing unit 20. The wavelength range of the spectral reflectance is 300 to 2500 nm.

[0033] In another embodiment, the wavelength range of the spectral reflectance is 900 to 1700 nm. When the wavelength range is 900 to 1700 nm, the test analyte 10 with different compositions has a more obvious difference in the spectral reflectance 54.

[0034] Referring again to Figures 2 and 3, as shown, in step S20, the image capturing unit 20 transmits the first information 50 to a control unit 40. In this embodiment, the image capturing unit 20 can be electrically connected directly to the control unit 40 to transmit the first information 50 to the control unit 40, or it can be connected and transmit data through wireless connection methods such as Bluetooth and wireless network. This invention is not limited to these methods.

[0035] Please refer to Figures 2 and 4. Figure 4 is a schematic diagram of the steps of the hyperspectral analysis method for blended materials of the present invention. As shown in the figure, in step S30, the control unit 40 performs a preprocessing on the first information 50 to generate a second information 60. In this embodiment, the control unit 40 can perform common preprocessing methods such as noise removal, category data processing, and data feature scaling on the first information 50 to enable the second information 60 generated after the preprocessing to converge faster and improve accuracy, or to supplement its values ​​to avoid it being unable to perform regression and classification.

[0036] Please refer to Figures 2 and 5. Figure 5 is a schematic diagram of the steps of the hyperspectral analysis method for blended materials of the present invention. As shown in the figure, in step S40, the control unit 40 uses a convolutional neural network 42 to obtain a plurality of feature values ​​62 of the second information 60. In this embodiment, the control unit 40 can perform common preprocessing methods such as noise removal, category data processing, and data feature scaling on the first information 50 to enable the second information 60 generated after the preprocessing to converge faster and improve accuracy, or to supplement its values ​​to avoid it being unable to perform regression and classification.

[0037] Referring again to Figures 2 and 5, as shown in the figures, in step S50, the control unit 40 judges the feature values ​​62 based on a deep learning model 44 to generate a judgment result 64. The judgment result 64 corresponds to the material composition of the test object 10. In this embodiment, the judgment result 64 corresponds to the ratio of polyester fiber to pure cotton.

[0038] Referring again to Figures 2 and 5, as shown in the figures, in step S60, the control unit 40 classifies the test object 10 according to the judgment result 64. The classification is based on the proportion of polyester fiber and pure cotton. In this embodiment, the classification divides the test object 10 into 7 categories, namely, polyester fiber 0% pure cotton 100%, polyester fiber 20% pure cotton 80%, polyester fiber 35% pure cotton 65%, polyester fiber 50% pure cotton 50%, polyester fiber 65% pure cotton 35%, polyester fiber 80% pure cotton 20%, and polyester fiber 100% pure cotton 0%.

[0039] Continuing from the above, the judgment result 64 generated by the hyperspectral analysis method of the blended material of the present invention is compared with the conventional material detection of the test object 10 using the chemical combustion method. The comparison shows that the correlation between the judgment result 64 generated by the hyperspectral analysis method of the blended material of the present invention and the result determined by the chemical combustion method is as high as 0.9790, and the accuracy of the quantitative analysis of material ratio is as high as 95%. This indicates that the hyperspectral analysis method of the blended material of the present invention has a very high reliability.

[0040] Continuing from the above, a practical example is provided here. A sample fabric of unknown proportion (the test object 10) is provided under the light source 30, and the optical image 52 and the spectral reflectance 54 are obtained through a hyperspectral image capturing device (the image capturing unit 20). The image is then transmitted to a host (the control unit 40), and the second information 60, which has been segmented and noise-removed, is generated through preprocessing. The host obtains the feature values ​​62 of the second information 60 through the convolutional neural network, and uses the deep learning model 44 to judge the feature values ​​62 to generate the comparison result 64. This allows the host to classify the sample fabric of unknown proportion, thereby determining the specific composition ratio of the sample fabric of unknown proportion.

[0041] Another embodiment is provided here. Please refer to Figure 6, which is a flowchart of the hyperspectral analysis method for blended materials of the present invention. Step S30 further includes steps S32 and S34, as shown in the figure. Since steps S10-S20 and S40-S60 have been described in the foregoing description, they will not be repeated here. The explanation is as follows:

[0042] Step S32: The control unit divides the first information to generate a plurality of segmented information;

[0043] Step S34: The control unit classifies the segmented information using a spectral angle matching algorithm and generates second information.

[0044] The following steps will be described in detail. Please refer to Figure 2 and Figure 7. Figure 7 is a schematic diagram of the steps of the hyperspectral analysis method for blended materials of the present invention. In step S32, the control unit 40 divides the first information 50 to generate a plurality of segmented information 56. In this embodiment, the first information 50 is approximately one optical image 52 of 640*1490 pixels. After division, it becomes the segmented information 56, which is approximately 232 images of 64*64 pixels. This provides better training data to the deep learning model 44 and allows the subsequent convolutional neural network 42 to better extract the feature values.

[0045] Referring again to Figures 2 and 7, as shown, in step S34, the control unit 40 classifies the segmented information 56 using a spectral angle matching algorithm and generates the second information 60. The judgment result 64 corresponds to the material composition of the test object 10. In this embodiment, the spectral angle matching algorithm calculates the angle between two spectral points, that is, the angle between these two points and the origin. By measuring the spectral angle difference, different materials can be classified, so that the subsequent convolutional neural network 42 can better extract these feature values.

[0046] The embodiments described above provide a hyperspectral analysis method for blended materials. An image acquisition unit acquires the optical image and spectral reflectance of the test object and transmits them to a control unit. The control unit preprocesses the optical image and spectral reflectance, obtains their feature values ​​through a convolutional neural network, and uses a deep learning model for judgment. In this way, the material composition of the test object can be detected by acquiring its optical data without damaging the object itself. Due to the large detection speed and area, it can be performed on a large scale and quickly, effectively improving the speed and quality of detection. It can also detect or assess the quality of all test objects with unknown compositions, eliminating the need for the unstable method of random sampling.

[0047] Therefore, this invention is indeed novel, inventive, and industrially applicable, and undoubtedly meets the requirements for patent application under the Patent Law of our country. Thus, we hereby file an invention patent application in accordance with the law, and earnestly pray that the Bureau will grant the patent as soon as possible.

[0048] However, the above description is only a preferred embodiment of the present invention and is not intended to limit the scope of the present invention. All equivalent changes and modifications made to the shape, structure, features and spirit described in the claims of the present invention should be included in the scope of the claims of the present invention.

[0049] S10-S60: Steps 1: Hyperspectral Analysis System for Blended Materials 10: Test Item 20: Image Capture Unit 30: Light source 40: Control Unit 42: Convolutional Neural Networks 44: Deep Learning Models 50: First Information 52: Optical Imaging 54: Spectral reflectance 56: Segmented Information 60: Second Information 62: Eigenvalues 64: Second Information

Claims

1. A hyperspectral analysis method for a blended material, comprising: acquiring first information of a test object using an image acquisition unit, the first information including an optical image and a spectral reflectance; transmitting the first information to a control unit; preprocessing the first information to generate second information; obtaining a plurality of feature values ​​of the second information using a convolutional neural network, the feature values ​​corresponding to the spectral reflectance of the test object; judging the feature values ​​based on a deep learning model to generate a judgment result, the judgment result corresponding to the ratio of polyester fiber to pure cotton in the test object; and classifying the ratio of polyester fiber to pure cotton in the test object based on the judgment result; wherein... The classification system is divided into multiple categories with the proportion of polyester fiber to pure cotton being 0% / 100%, 20% / 80%, 35% / 65%, 50% / 50%, 65% / 35%, 80% / 20%, and 100% / 0%.

2. The hyperspectral analysis method for blended materials as described in claim 1, wherein the step of the control unit preprocessing the first information to generate a second information further includes the steps of: the control unit segmenting the first information to generate a plurality of segmented information; and the control unit classifying the segmented information through a spectral angle matching algorithm to generate the second information.

3. The hyperspectral analysis method for blended materials as described in claim 1, wherein in the step of acquiring first information of a analyte using an image capturing unit, the first information comprising an optical image and a spectral reflectance, the wavelength range of the spectral reflectance is 900 nanometers to 1700 nanometers.

Images