A method, system and device for identifying moldy wheat based on terahertz imaging

Abstract

This invention proposes a method, system, and apparatus for identifying moldy wheat based on terahertz imaging. The method includes: Step 1, obtaining a three-dimensional terahertz image containing spatial and spectral information; Step 2, acquiring frequency domain image data through a terahertz reflection imaging system; performing three-dimensional thermal imaging by accumulating Z frequency domain image data corresponding to each pixel in the frequency domain image, and dividing the moldy wheat into regions of interest using the amplitude value T; after determining the regions of interest, extracting the spectral information corresponding to each pixel based on the coordinate position, and averaging the spectrum of each grain of wheat; Step 3, based on the obtained spectral average value, classifying and identifying wheat at different mold growth times using an improved VGG19-Inception-ResNet-A network, obtaining frequency domain spectral curves for different degrees of mold; and determining the degree of mold in the wheat based on the amplitude value of the frequency domain spectral curve.

Description

Technical Field

[0001] This invention relates to a method for identifying moldy wheat, and more particularly to a method, system, and apparatus for identifying moldy wheat based on terahertz imaging. Background Technology

[0002] Early mold growth in wheat is difficult to detect, leading to progressively worsening mold and significant losses during storage. Timely and accurate identification of moldy wheat is crucial for ensuring food security. Traditional methods for detecting wheat mold include sensory evaluation, fluorescence staining, and mass spectrometry coupled with gas chromatography. While these methods offer high accuracy, they are time-consuming and complex. Currently, common methods such as electronic nose technology, infrared spectroscopy, and hyperspectral imaging effectively address the issues of long processing times and complex operations associated with traditional techniques. However, these methods also have limitations. For example, electronic nose technology is susceptible to sensor sensitivity and environmental factors, has low equipment integration, is expensive, and has a short lifespan. Infrared spectroscopy has relatively low sensitivity, is easily affected by environmental factors, and is highly dependent on chemical calculations. Hyperspectral imaging involves massive amounts of data and complex image processing. Furthermore, all these methods share a common limitation: they cannot achieve "see-through" imaging of the object under non-destructive conditions, meaning they cannot achieve internal imaging of the object without damage.

[0003] Terahertz waves (THz) are electromagnetic waves between microwaves and infrared, with frequencies between 0.1 and 10 THz. They have the characteristics of strong penetration, low ionizing radiation, and molecular fingerprinting. They also have advantages such as fast detection and analysis speed, simultaneous multi-component analysis, and non-destructive testing, which can effectively make up for the shortcomings of traditional detection methods and currently common technologies.

[0004] Therefore, a method for identifying moldy wheat based on terahertz imaging technology was designed and studied to achieve accurate and rapid identification of moldy wheat, improve the efficiency of mold detection, and ensure the safety of stored grain quality. This method has practical significance and good application prospects. Summary of the Invention

[0005] To address the aforementioned issues, it is necessary to provide a method, system, and apparatus for identifying moldy wheat based on terahertz imaging.

[0006] The first aspect of this invention provides a method for identifying moldy wheat based on terahertz imaging, comprising:

[0007] Step 1, Image Acquisition

[0008] Wheat samples with different mold growth times were placed on a moving platform of a terahertz imaging system and scanned for reflection imaging to obtain three-dimensional terahertz images containing spatial and spectral information.

[0009] Step 2, spectral extraction

[0010] Acquire frequency domain image data using a terahertz reflection imaging system;

[0011] After accumulating the Z frequency domain image data corresponding to each pixel in the frequency domain image, a three-dimensional thermal image is formed. The region of interest is divided into regions of interest for moldy wheat using the amplitude value T. The spectral range with an amplitude value less than T is taken as the region of interest for wheat, and all spectral ranges with an amplitude value greater than or equal to T are set to 0.

[0012] After determining the region of interest, the spectral information corresponding to each pixel is extracted based on its coordinate position, and the average value of the spectrum of each grain of wheat is taken.

[0013] Step 3, Mold Identification

[0014] Based on the obtained spectral average values, the improved VGG19-Inception-ResNet-A network was used to classify and identify wheat at different mold growth times, and frequency domain spectral curves of different mold growth degrees were obtained.

[0015] The degree of mold growth in wheat can be determined by the amplitude value of the frequency domain spectral curve. The more severe the mold growth, the larger the peak amplitude value between 0-0.5 THz.

[0016] The second aspect of the present invention provides a terahertz imaging-based system for identifying moldy wheat, comprising: an image acquisition module for acquiring a three-dimensional terahertz image of moldy wheat containing spatial and spectral information placed on a moving platform of a terahertz imaging system;

[0017] The spectral extraction module, connected to the image acquisition module, is used for extracting spectral information. The extraction of spectral information is implemented using the following method:

[0018] Acquire frequency domain image data using a terahertz reflection imaging system;

[0019] After accumulating the Z frequency domain image data corresponding to each pixel in the frequency domain image, a three-dimensional thermal image is formed. The region of interest is divided into regions of interest for moldy wheat using the amplitude value T. The spectral range with an amplitude value less than T is taken as the region of interest for wheat, and all spectral ranges with an amplitude value greater than or equal to T are set to 0.

[0020] After determining the region of interest, the spectral information corresponding to each pixel is extracted based on its coordinate position, and the average value of the spectrum of each grain of wheat is taken.

[0021] The mold detection module, connected to the spectral extraction module, is used to identify the degree of mold growth in wheat. The identification of the degree of mold growth in wheat is achieved using the following method:

[0022] Based on the obtained spectral average values, the improved VGG19-Inception-ResNet-A network was used to classify and identify wheat at different mold growth times, and frequency domain spectral curves of different mold growth degrees were obtained.

[0023] The degree of mold growth in wheat can be determined by the amplitude value of the frequency domain spectral curve. The more severe the mold growth, the larger the peak amplitude value between 0-0.5 THz.

[0024] A third aspect of the present invention provides a device for identifying moldy wheat, comprising:

[0025] Memory; and

[0026] A processor coupled to the memory is configured to execute the terahertz imaging-based method for identifying moldy wheat based on instructions stored in the memory.

[0027] A fourth aspect of the present invention provides a non-transient computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for identifying moldy wheat based on terahertz imaging.

[0028] This invention has outstanding substantive features and significant progress compared to the prior art, specifically:

[0029] This invention utilizes terahertz imaging technology to obtain spectral data of wheat and combines it with an improved VGG19-Inception-ResNet-A network to classify and identify wheat samples at different mold growth stages. The improved VGG19-Inception-ResNet-A network achieves higher classification accuracy than VGG19, and the network testing time is shorter than that of Inception-ResNet-V2.

[0030] Additional aspects and advantages of the invention will become apparent in the following description or may be learned by practice of the invention. Attached Figure Description

[0031] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0032] Figure 1 This is a flowchart of the identification process of the present invention.

[0033] Figure 2 The present invention uses FV software to display time-domain (a) and frequency-domain (b) images of moldy wheat.

[0034] Figure 3 This is a hypercube image of moldy wheat according to the present invention.

[0035] Figure 4 This is a three-dimensional image of moldy wheat according to the present invention.

[0036] Figure 5 This is a spectrum of wheat mold growth at different degrees.

[0037] Figure 6 This is a distribution map of PLSR samples.

[0038] Figure 7 This is a sample error distribution diagram of PLSR.

[0039] Figure 8 This is the network architecture diagram of VGG19-Inception-ResNet-A. Detailed Implementation

[0040] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0041] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0042] Example 1

[0043] like Figure 1 As shown, this embodiment proposes a method for identifying moldy wheat based on terahertz imaging, including:

[0044] Step 1, Image Acquisition

[0045] Wheat samples with different mold growth times were placed on a mobile platform of a terahertz imaging system and scanned for reflectance imaging to obtain three-dimensional terahertz images containing spatial and spectral information. Specifically, the maximum scanning area of ​​the mobile platform of the terahertz imaging system is 50mm×50mm, and the spatial resolution is 0.1mm. In the three-dimensional terahertz image, each pixel corresponds to a terahertz spectrum, and each spectrum contains 375 time-domain points, corresponding to a time-domain range of 0~20ps and a frequency-domain range of 0~5THz.

[0046] Step 2, spectral extraction

[0047] Frequency domain image data was acquired using a terahertz reflection imaging system, and the wheat frequency domain image data was imaged using FV software to obtain an image format, such as... Figure 2 As shown;

[0048] Because spectral data is in a three-dimensional data format, such as Figure 3 As shown in (a), three-dimensional heat map imaging is performed after accumulating Z frequency domain image data corresponding to each pixel in the frequency domain image. The region of interest is divided into regions of interest for moldy wheat using the amplitude value T. The spectral range with an amplitude value less than T is taken as the region of interest for wheat, and all regions with an amplitude value greater than or equal to T are set to 0. In the three-dimensional heat map, the wheat region and the background region have significantly different shapes and colors. The amplitude value T is selected based on the amplitude value corresponding to the point of change in shape and color.

[0049] After determining the region of interest, the spectral information corresponding to each pixel is extracted based on its coordinate position, and the average value of the spectrum of each grain of wheat is taken.

[0050] In this step, Z frequency domain image data are accumulated and then used for three-dimensional thermal imaging. This involves merging Z spectral image data into a single wheat frequency domain image and then imaging it. The result is as follows: Figure 3 As shown in (b), it can be seen that the color of the wheat area differs from the background color. This phenomenon can be analyzed using a three-dimensional heatmap, as follows: Figure 4 (a) shows that the amplitude of the wheat sample is smaller than that of the background, which is mainly due to the absorption of the terahertz reflection spectrum in reflection imaging, resulting in a reduced spectral amplitude and a darker color in the image. Based on the amplitude value, the region of interest (ROI) for the wheat can be selected. After determining the ROI, the spectral information corresponding to each pixel is extracted based on its coordinate position, and the average value of the spectrum for each wheat grain is taken. Figure 5 As shown, the differences in different degrees of mold growth can be seen from the frequency domain spectral curves.

[0051] Step 3, Mold Identification

[0052] Based on the obtained spectral average values, the improved VGG19-Inception-ResNet-A network was used to classify and identify wheat at different mold growth times, and frequency domain spectral curves of different mold growth degrees were obtained.

[0053] The degree of mold growth in wheat can be determined by the amplitude value of the frequency domain spectral curve. The more severe the mold growth, the larger the peak amplitude value between 0-0.5 THz.

[0054] In this step, such as Figure 8As shown, the improved VGG19-Inception-ResNet-A network comprises an input layer, unit block A, an Inception-resnet-A block, unit block B, a fully connected layer, and an output layer connected in sequence. Unit blocks A and B each include convolutional and pooling layers connected in sequence. The Inception-resnet-A block includes three convolutional and pooling operations of different sizes, each implemented by a convolutional and pooling layer connected in sequence. Specifically, for the VGG19-Inception-ResNet-A network, the input data is a one-dimensional array of 500*1*4830. After convolution and pooling operations in unit block A, the output data is 250*1*384. The output data is then convolved with a 1*1 convolution before being input into the Inception-resnet-A module. After three convolutional and pooling operations of different sizes, the output data is again 250*1*384. The data from unit A is then fused with the data from the Inception-ResNet-A module and input into unit B. After convolution and pooling operations in unit B, the output data is 16*1*512. After the first two fully connected operations, the data is 4096*1. Finally, the Softmax function is used in the last fully connected layer for tri-class classification.

[0055] In this embodiment, the color change is significant at an amplitude value of 1900 a.u. Therefore, the spectral range with an amplitude value T less than 1900 a.u is selected as the region of interest for wheat, while spectral values ​​greater than or equal to 1900 a.u are all set to 0. Figure 4 As shown in (b);

[0056] Because using amplitude values ​​as the region of interest (ROI) for dividing moldy wheat grains is random, it can easily lead to the selection of wheat grain regions being too large or too small. Both excessively large and small selections result in a lack of representativeness in the selected spectra. Therefore, to verify the effectiveness of the selected spectral data (spectral data extracted from the ROI region selected by amplitude value T), PLSR regression analysis was performed on the selected moldy wheat spectra and background spectra. Before the regression analysis, labels 1 and 2 were added to the moldy wheat spectra and background spectra, respectively, i.e., Y values ​​were set to 1 and 2, respectively. The `train_test_split` function was used to divide all spectral datasets into training and test sets in a 7:3 ratio. The results show that the test set results obtained using PLSR analysis are not significantly different from the true values. Figure 6 It can be seen that the test set samples are mostly distributed around y=1 and y=2. To accurately obtain the error range of each test set sample, the predicted value is subtracted from the actual value, and the result is as follows. Figure 7It can be seen that the difference between the predicted value and the true value is mostly distributed between 0.2 and -0.2. Through the above PLSR analysis, it can be seen that there is a clear distinction between the selected wheat spectrum and the background spectrum, that is, the selected wheat grain range (the determined region of interest) and spectrum (the spectrum extracted from the region of interest selected by the amplitude value T) are effective and can be used for wheat spectrum classification and identification.

[0057] Analysis of Experimental Results

[0058] To compare the effectiveness of the improved algorithm, frequency domain spectral data were input into the VGG19, Inception-ResNet-V2, and VGG19-Inception-ResNet-A network models, respectively, and their corresponding loss values, accuracy, precision, recall, and test time were obtained. Specific data are shown in Table 1. It can be seen that the test set accuracy of all three networks can reach over 99%, but the significant difference lies in the test time. The 1D-Inception-ResNet-V2 network took the longest, at 321 seconds, with an accuracy of 99.77%; the 1D-VGG19 network took the shortest, at 77 seconds, but its accuracy was lower than that of the 1D-Inception-ResNet-V2 network. The network based on 1D-VGG19 and 1D-Inception-ResNet-V2 and the improved ID-VGG19-Inception-ResNet-A combines the advantages of both networks, improving accuracy by 0.35% compared to the 1D-VGG19 model and reducing training time by 226 seconds compared to 1D-Inception-ResNet-V2.

[0059] To achieve accurate and rapid identification of different types of moldy wheat, and to demonstrate the effectiveness of the improved algorithm in this embodiment, three networks—1D-VGG19, 1D-Inception-Renest-v2, and 1D-VGG19-Inception-ResNet-A (an improvement on VGG19)—were used as methods for identifying moldy wheat. The experimental dataset consisted of 4830 data entries, divided into training and test sets in a 7:3 ratio. To ensure the networks operated under the same variable environment, the learning rate was 0.00001, the number of epochs was 20, the cross-entropy loss function was used, and the Adam optimizer was employed. All experiments were conducted on the same server.

[0060] Table 1 Comparison of Experimental Results

[0061] network Loss value (%) Accuracy (%) Precision (%) Recall(%) Test duration (s) VGG19 3.9 99.11 99.11 99.10 77 Inception-Resnet-V2 1.01 99.77 99.77 99.77 321 VGG19-Inception-Resnet-A 1.85 99.45 99.58 99.80 95

[0062] Example 2

[0063] This embodiment provides a terahertz imaging-based system for identifying moldy wheat, including: an image acquisition module for acquiring a three-dimensional terahertz image of moldy wheat containing spatial and spectral information placed on a mobile platform of the terahertz imaging system;

[0064] The spectral extraction module, connected to the image acquisition module, is used for extracting spectral information. The extraction of spectral information is implemented using the following method:

[0065] Acquire frequency domain image data using a terahertz reflection imaging system;

[0066] After accumulating the Z frequency domain image data corresponding to each pixel in the frequency domain image, a three-dimensional thermal image is formed. The region of interest is divided into regions of interest for moldy wheat using the amplitude value T. The spectral range with an amplitude value less than T is taken as the region of interest for wheat, and all spectral ranges with an amplitude value greater than or equal to T are set to 0.

[0067] After determining the region of interest, the spectral information corresponding to each pixel is extracted based on its coordinate position, and the average value of the spectrum of each grain of wheat is taken.

[0068] The mold detection module, connected to the spectral extraction module, is used to identify the degree of mold growth in wheat. The identification of the degree of mold growth in wheat is achieved using the following method:

[0069] Based on the obtained spectral average values, the improved VGG19-Inception-ResNet-A network was used to classify and identify wheat at different mold growth times, and frequency domain spectral curves of different mold growth degrees were obtained.

[0070] The degree of mold growth in wheat can be determined by the amplitude value of the frequency domain spectral curve. The more severe the mold growth, the larger the peak amplitude value between 0-0.5 THz.

[0071] The specific identification method of the identification system in this embodiment refers to the method described in Embodiment 1, and will not be repeated here.

[0072] Example 3

[0073] This embodiment provides an electrocardiogram characteristic wave segmentation device, including:

[0074] Memory; and

[0075] A processor coupled to the memory is configured to execute the terahertz imaging-based method for identifying moldy wheat as described in Example 1, based on instructions stored in the memory.

[0076] The memory may include, for example, system memory, fixed non-volatile storage media, etc. System memory may store, for example, the operating system, application programs, boot loader, and other programs.

[0077] The device may also include input / output interfaces, network interfaces, and storage interfaces. These interfaces, as well as the memory and processor, can be connected via, for example, a bus. The input / output interfaces provide connection interfaces for input / output devices such as monitors, mice, keyboards, and touchscreens. The network interfaces provide connection interfaces for various networked devices. The storage interfaces provide connection interfaces for external storage devices such as SD cards and USB flash drives.

[0078] Example 4

[0079] This embodiment provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the terahertz imaging-based method for identifying moldy wheat as described in Embodiment 1.

[0080] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-non-transitory readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer program code.

[0081] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0082] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0083] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0084] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying moldy wheat based on terahertz imaging, characterized in that, include: Step 1, Image Acquisition Wheat samples with different mold growth times were placed on a moving platform of a terahertz imaging system and scanned for reflection imaging to obtain three-dimensional terahertz images containing spatial and spectral information. In a three-dimensional terahertz image, each pixel corresponds to a terahertz spectrum, and each spectrum contains 375 time-domain points, corresponding to a time-domain range of 0~20ps and a frequency-domain range of 0~5THz. Step 2, spectral extraction Acquire frequency domain image data using a terahertz reflection imaging system; After accumulating Z frequency domain image data corresponding to each pixel in the frequency domain image, a three-dimensional heat map is formed. The amplitude value T is used to divide the moldy wheat into regions of interest. The spectral range with an amplitude value less than T is the wheat region of interest, and all amplitude values ​​greater than or equal to T are set to 0. The method for selecting the amplitude value T is as follows: in the three-dimensional heat map, the wheat region and the background region have significantly different shapes and colors. The amplitude value corresponding to the point of change in shape and color is used as the basis for selecting the amplitude value T. After determining the region of interest, the spectral information corresponding to each pixel is extracted based on the coordinate position of the region of interest, and the average value of the spectrum of each grain of wheat is taken. Step 3, Mold Identification Based on the obtained spectral average values, the improved VGG19-Inception-ResNet-A network was used to classify and identify wheat at different mold growth times, and frequency domain spectral curves of different mold growth degrees were obtained. The degree of mold growth in wheat can be determined by the amplitude value of the frequency domain spectral curve. The more severe the mold growth, the larger the peak amplitude value between 0-0.5 THz. The improved VGG19-Inception-ResNet-A network consists of an input layer, cell block A, an Inception-resnet-A block, cell block B, a fully connected layer, and an output layer connected in sequence. The input layer receives a 500*1*4830 one-dimensional array composed of the average spectral values; both unit block A and unit block B include sequentially connected convolutional layers and pooling layers for dimensionality reduction and feature extraction of the input data; the Inception-ResNet-A block includes three convolutional and pooling operations of different sizes, each implemented by sequentially connected convolutional and pooling layers, for parallel extraction of multi-scale features; the data output from unit block A is input into the Inception-ResNet-A block after 1*1 convolution, and the output data of unit block A is fused with the output data of the Inception-ResNet-A block before being input into unit block B.

2. The method for identifying moldy wheat based on terahertz imaging according to claim 1, characterized in that, The amplitude value T is 1900a.u.

3. A system for identifying moldy wheat based on terahertz imaging, characterized in that, include: The image acquisition module is used to acquire a three-dimensional terahertz image containing spatial and spectral information of moldy wheat placed on the mobile platform of the terahertz imaging system. In the three-dimensional terahertz image, each pixel corresponds to a terahertz spectrum, and each spectrum contains 375 time-domain points, corresponding to a time-domain range of 0~20ps and a frequency-domain range of 0~5THz. The spectral extraction module, connected to the image acquisition module, is used for extracting spectral information. The extraction of spectral information is implemented using the following method: Acquire frequency domain image data using a terahertz reflection imaging system; After accumulating Z frequency domain image data corresponding to each pixel in the frequency domain image, a three-dimensional heat map is formed. The amplitude value T is used to divide the moldy wheat into regions of interest. The spectral range with an amplitude value less than T is the wheat region of interest, and all amplitude values ​​greater than or equal to T are set to 0. The method for selecting the amplitude value T is as follows: in the three-dimensional heat map, the wheat region and the background region have significantly different shapes and colors. The amplitude value corresponding to the point of change in shape and color is used as the basis for selecting the amplitude value T. After determining the region of interest, the spectral information corresponding to each pixel is extracted based on its coordinate position, and the average value of the spectrum of each grain of wheat is taken. The mold detection module, connected to the spectral extraction module, is used to identify the degree of mold growth in wheat. The identification of the degree of mold growth in wheat is achieved using the following method: Based on the obtained spectral average values, the improved VGG19-Inception-ResNet-A network was used to classify and identify wheat at different mold growth times, and frequency domain spectral curves of different mold growth degrees were obtained. The degree of mold growth in wheat can be determined by the amplitude value of the frequency domain spectral curve. The more severe the mold growth, the larger the peak amplitude value between 0-0.5 THz. The improved VGG19-Inception-ResNet-A network consists of an input layer, cell block A, an Inception-resnet-A block, cell block B, a fully connected layer, and an output layer connected in sequence. The input layer receives a 500*1*4830 one-dimensional array composed of the average spectral values; both unit block A and unit block B include sequentially connected convolutional layers and pooling layers for dimensionality reduction and feature extraction of the input data; the Inception-ResNet-A block includes three convolutional and pooling operations of different sizes, each implemented by sequentially connected convolutional and pooling layers, for parallel extraction of multi-scale features; the data output from unit block A is input into the Inception-ResNet-A block after 1*1 convolution, and the output data of unit block A is fused with the output data of the Inception-ResNet-A block before being input into unit block B.

4. The terahertz imaging-based moldy wheat identification system according to claim 3, characterized in that, The amplitude value T is 1900a.u.

5. A device for identifying moldy wheat, comprising: Memory; as well as A processor coupled to the memory, the processor being configured to execute, based on instructions stored in the memory, the terahertz imaging-based method for identifying moldy wheat as described in any one of claims 1-2.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for identifying moldy wheat based on terahertz imaging as described in any one of claims 1-2.

Images