Oil product identification method, device and computer equipment based on improved LetNet network

Abstract

The present invention provides an oil product identification method, device, and computer equipment based on an improved LetNet network. The method comprises: obtaining fluorescence spectral images of multiple different oil products; preprocessing each fluorescence spectral image to obtain preprocessed image data; improving a traditional LetNet network to obtain an improved LetNet network; constructing a training set based on the preprocessed image data and training the improved LetNet network, thereby obtaining a trained oil product identification model; and using the trained oil product identification model to identify the preprocessed fluorescence spectral image of the oil product to be identified, outputting an oil product identification classification result. The present invention combines a deep learning algorithm with fluorescence spectral data and improves the network structure of the LetNet network, reducing computational complexity, enhancing model fitting capabilities, and improving the accuracy of oil product identification.

Description

Technical Field

[0001] The present invention relates to the technical field of oil spill identification, and in particular to an oil product identification method, device and computer equipment based on an improved LetNet network. Background Art

[0002] As the lifeblood of industry, oil continues to fuel the rapid growth of the global industrial economy, and demand for oil energy is also rapidly increasing worldwide. Typically, oil is transported by sea. As oil transportation becomes increasingly frequent, it can eventually leak into the ocean, wasting human and material resources and posing a serious threat to the ecological environment. Therefore, obtaining timely and accurate information about the spilled oil after an oil spill is crucial for the subsequent management and restoration of the marine ecosystem.

[0003] Early methods for identifying oil spills included turbidity, gravimetry, and suspension methods. These methods primarily differentiated oil types based on the specific physical properties of oil-water solutions. However, oil spills in seawater are affected by the natural environment, undergoing evaporation, weathering, sedimentation, and dissolution, which alter their physical and chemical properties and pose significant challenges for identifying oil spill types. These methods all suffer from limitations such as limited application scope and low accuracy in identifying oil spill types. With continuous technological advancements, advanced techniques such as spectroscopy, chromatography, and mass spectrometry have been gradually applied to identify marine oil spill types, achieving excellent results. Fluorescence spectroscopy, with its strong anti-interference capabilities, high accuracy, and high sensitivity, has gradually become the mainstream method for identifying oil spills.

[0004] In recent years, with the continuous efforts of scientific researchers in our country, a lot of work has been done in the application field of fluorescence detection technology, and certain research results have been achieved. In 2010, Tian Guangjun used a combination of neural networks and singular value decomposition to extract features and reconstruct the three-dimensional fluorescence spectrum of oil spills, and successfully identified different types of oil spills. In 2013, Yang Lili, Wang Yutian and others proposed a method for detecting oil spill pollutants that combines parallel factor analysis algorithm with fluorescence spectroscopy technology. This method has a certain anti-interference ability and can accurately identify the types and predict the concentrations of petroleum pollutant samples in complex unknown environments. In 2020, Jin Yun built an oil spill identification model based on the YOLOv3 algorithm framework, which achieved accurate identification and positioning of oil spills. In 2021, Jiao Huihui used a combination of principal component analysis and characteristic parameters to extract features from fluorescence spectral data, and then used least squares discriminant analysis and BP neural network to complete the identification of oil spill types.

[0005] Researchers at home and abroad mainly use fluorescence spectroscopy to analyze and identify oil spill types, and manually extract the characteristics of the oil fluorescence spectrum. This process has the disadvantages of consuming a lot of manpower and material resources and low work efficiency. It cannot effectively avoid the influence of noise signals on fluorescence spectrum data, which leads to deviations in identification results. Deep learning algorithms use neural networks to efficiently extract feature information of image data, reducing the impact of noise signals on detection results. Summary of the Invention

[0006] In order to solve the above technical problems, the present invention combines deep learning algorithms with fluorescence spectroscopy and proposes an oil product identification method, device and computer equipment based on an improved LetNet network. The noise signal of the oil product fluorescence spectrum image is removed by filtering, and then the oil product fluorescence spectrum image dataset is input into the network for training. The oil spill is accurately identified based on the significant features of the oil product fluorescence spectrum.

[0007] According to a first aspect of the present invention, the present invention provides an oil product identification method based on an improved LetNet network, comprising the following steps:

[0008] Obtain fluorescence spectrum images of various oil products;

[0009] Preprocessing each of the fluorescence spectrum images to obtain preprocessed image data;

[0010] Improve the traditional LetNet network to obtain an improved LetNet network;

[0011] Building a training set based on the preprocessed image data, training the improved LetNet network, and obtaining a trained oil product recognition model after the training is completed;

[0012] The pre-processed fluorescence spectrum image of the oil product to be identified is identified by the trained oil product identification model, and the oil product identification classification result is output.

[0013] Preferably, the step of obtaining fluorescence spectrum images of a plurality of different oil products includes:

[0014] The ultraviolet fluorescence method was used to obtain fluorescence spectrum images of various oil products.

[0015] Preferably, the pre-processing step includes:

[0016] The fluorescence spectrum image is filtered using an FIR filter.

[0017] Preferably, the pre-processing step further comprises:

[0018] Input the fluorescence spectrum image after filtering, first perform channel conversion on the image to remove the transparent channel in the image, then transform the image to change the size of the original image, randomly rotate the image, and center crop the image. After completing the above steps, convert the image into Tensor image data, then perform normalization operation, grayscale operation on the image, add a dimension to the Tensor image data, and finally output the preprocessed image data.

[0019] Preferably, the improved LetNet network includes:

[0020] The sequentially connected input layer, first convolution layer, first activation function layer, second convolution layer, first pooling layer, second activation function layer, third convolution layer, third activation function layer, second pooling layer, fully connected layer and output layer;

[0021] The first activation function layer, the second activation function layer, and the third activation function layer are used to enhance the nonlinear fitting ability of the network;

[0022] In the process of full connection of the fully connected layer, dropout is used to inactivate neurons with a probability of 0.25, and the flatten function, linear function and log_softmax function are used to complete the recognition function of the network.

[0023] Preferably, in the improved LetNet network structure, the convolution kernel size used in the three convolution layers is 3*3.

[0024] Preferably, the step of identifying the pre-processed fluorescence spectrum image of the oil product to be identified by using the trained oil product identification model and outputting the oil product identification classification result includes:

[0025] Preprocess the fluorescence spectrum image of the oil product to be identified;

[0026] The preprocessed fluorescence spectrum image of the oil product to be identified is input into the trained oil product identification model. After convolution calculation by the first convolution layer, it is fitted by the first activation function layer, and then convolution calculation by the second convolution layer. The output result is subjected to maximum pooling processing by the first pooling layer, fitted by the second activation function layer, convolution calculation by the third convolution layer, fitted by the third activation function layer, and maximum pooling processing by the second pooling layer. Dropout processing is then performed, the flatten function is used to flatten the multidimensional matrix, the Linear function is used for full connection, and finally the log_softmax function is used to complete the recognition of the oil product image.

[0027] Preferably, after the step of outputting the oil product identification and classification result, the method further includes:

[0028] The oil product identification and classification results were evaluated by overall accuracy and confusion matrix.

[0029] According to a second aspect of the present invention, the present invention provides an oil product identification device based on an improved LetNet network, comprising the following modules:

[0030] Image acquisition module, used to obtain fluorescence spectrum images of various oil products;

[0031] An image preprocessing module, used to preprocess each of the fluorescence spectrum images to obtain preprocessed image data;

[0032] The network improvement module is used to improve the traditional LetNet network to obtain an improved LetNet network;

[0033] A network training module is used to construct a training set based on the preprocessed image data and train the improved LetNet network. After the training is completed, a trained oil product recognition model is obtained;

[0034] The oil product identification module is used to identify the pre-processed fluorescence spectrum image of the oil product to be identified through the trained oil product identification model and output the oil product identification classification result.

[0035] According to a third aspect of the present invention, the present invention provides a computer device, which includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the oil product identification method is executed.

[0036] The technical solution provided by the present invention has the following beneficial effects:

[0037] (1) Compared with the LetNet network structure, the improved LetNet network structure uses ReLU instead of Sigmoid to improve convergence speed. The 5*5 convolution kernel used in the LetNet network is replaced with a smaller 3*3 convolution kernel. The use of a small convolution kernel reduces the network's computational workload to a certain extent. The improved LetNet network acquires more texture features of oil product images by adding multiple activation function layers and prevents network overfitting by adding dropout. The final results show that the improved LetNet network has a recognition rate of up to 94%.

[0038] (2) The improved LetNet network can more accurately extract key features of oil fluorescence spectra, such as the location and size of the fluorescence peak, and the profile of the fluorescence spectrum curve, thereby providing effective data support for the improved LetNet network to accurately identify oil images. Due to the addition of activation function layers in the improved LetNet network, the fitting ability of the network model is enhanced, the extracted feature information is clearer, and the recognition accuracy of the network is further improved.

[0039] (3) Compared with the VGG-16 network structure, the improved LetNet network structure model is streamlined. It occupies less running memory during calculations, which reduces the consumption of computing resources. Due to the different sizes of the input network images, the improved LetNet network needs to connect fewer nodes when performing full connection. The final results show that the recognition rate of the improved LetNet network is 4% higher than that of VGG-16. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The specific effects of the present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0041] Figure 1 This is an overall flow chart of an oil product identification method based on an improved LetNet network according to an embodiment of the present invention;

[0042] Figure 2 is a flow chart of pre-processing after filtering according to an embodiment of the present invention;

[0043] Figure 3 is a structural diagram of the improved LetNet network according to an embodiment of the present invention;

[0044] Figure 4 is a confusion matrix with a training ratio of 90% under the oil product image dataset according to an embodiment of the present invention;

[0045] Figure 5 This is a structural diagram of an oil product identification device based on an improved LetNet network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0047] Example 1:

[0048] Please refer to Figure 1 In this embodiment, an oil product identification method based on an improved LetNet network includes the following steps:

[0049] S1: Acquire fluorescence spectrum images of various oil products;

[0050] Specifically, the ultraviolet fluorescence method is used to obtain fluorescence spectrum images of different oil products, such as fluorescence spectrum images of lubricating oil, engine oil, transformer oil and motorcycle oil.

[0051] S2: preprocessing each of the fluorescence spectrum images to obtain preprocessed image data;

[0052] In this embodiment, the filter used is an FIR filter. Since there are a lot of high-frequency noise signals in the fluorescence spectrum data collected by the device based on the ultraviolet fluorescence method, the present invention uses a low-pass filter to filter the data. The specific process is as follows:

[0053] First, the acquired wavelength data is set as the horizontal axis of the fluorescence spectrum image. Then, this fluorescence wavelength data is Fourier transformed to obtain its corresponding frequency components. Finally, the parameters of the low-pass filter are designed based on these frequency components to complete the filtering operation of the fluorescence spectrum image. A total of 3648 wavelength data are collected. After the wavelength data is Fourier transformed, the parameters of the FIR low-pass filter selected are as follows: sampling frequency fs = 2048 Hz, pass frequency Fpass = 30 Hz, cutoff frequency Fstop = 200 Hz, and filter order N = 17.

[0054] Create an img folder and place the filtered fluorescence spectra images of lubricating oil, engine oil, transformer oil, and motorcycle oil into the corresponding oil folders. Then, place these folders containing the different oil products into the created img folder. Run create_tfrecords.py to convert the data into a format suitable for training the improved LetNet network. This will generate the train.tfrecords file, which contains the fluorescence spectra data for the four oil products and their corresponding labels.

[0055] refer to Figure 2 , Figure 2 This example preprocesses the filtered image using the averaging method. First, the filtered oil fluorescence spectrum image is input. Next, the image undergoes channel conversion to remove the transparent channel. The image is then transformed to resize the original image, randomly rotated, and center-cropped. After completing these steps, the image is converted to Tensor image data, normalized, grayscaled, and a dimension added to the Tensor image data. Finally, the preprocessed image data is output.

[0056] S3: Improve the traditional LetNet network to obtain an improved LetNet network;

[0057] refer to Figure 3 , Figure 3 The improved LetNet network structure is a novel network structure. The present invention optimizes and improves the original structure of the LetNet network. Compared with the traditional LetNet network structure, the improved LetNet network first adds multiple activation function layers to the network structure to enhance the nonlinear fitting ability of the network. Secondly, in the process of full connection, the LetNet network directly connects all neurons, while the improved LetNet network uses dropout to inactivate neurons with a probability of 0.25, and then uses the flatten function, the Linear function and the log_softmax function to complete the recognition function of the network. On the one hand, it can reduce the overall parameter amount, and on the other hand, it can effectively prevent the overfitting of the network model. In addition, considering the influence of the convolution kernel size on the parameters of the improved LetNet network, as shown in Table 1, the convolution kernel size used in the convolution layer of the improved LetNet network structure is set to 3*3, while the convolution kernel size used in the convolution layer of the traditional LetNet network structure is 5*5. Therefore, the improved LetNet network reduces the overall computational complexity of the network to a certain extent.

[0058] Table 1. The impact of convolution kernel size on the improved LetNet network parameters

[0059]

[0060] The parameter settings of each layer of the improved LetNet network are shown in Table 2:

[0061] Table 2 Parameters of each layer of the improved LetNet network model

[0062]

[0063] The input of the improved LetNet network training is a grayscale oil fluorescence spectrum image of size 224*224*1. The preprocessed oil fluorescence spectrum image is input into the improved LetNet network. After the convolution calculation of the first convolution layer Conv1, it is fitted by the first activation function layer ReLU1, and then convolution calculation by the second convolution layer Conv2. The output result is processed by the first pooling layer Maxpool1 maximum pooling, fitted by the second activation function layer ReLU2, and convolution calculation by the third convolution layer Conv3. It is fitted by the third activation function layer ReLU3, and the output result is processed by the second pooling layer Maxpool2 maximum pooling, and then dropout processing is performed. The flatten function is used to flatten the multidimensional matrix, and Linear is used for full connection. Finally, the log_softmax function is used to complete the recognition of the oil image.

[0064] S4: constructing a training set based on the preprocessed image data, training the improved LetNet network, and obtaining a trained oil product recognition model after the training is completed;

[0065] In step S4, a training set is constructed based on the preprocessed image data. In addition, there is a validation set, which does not require preprocessing.

[0066] S5: Identify the pre-processed fluorescence spectrum image of the oil product to be identified using the trained oil product identification model, and output the oil product identification classification result.

[0067] In step S5, the preprocessing process of the fluorescence spectrum image of the oil product to be identified is the same as step S2.

[0068] After outputting the oil product identification and classification results, this embodiment also performs an accuracy evaluation on the classification results.

[0069] In image classification tasks, the overall accuracy (OA) and confusion matrix (CM) are the two most widely used metrics for network evaluation. Therefore, in this example, the confusion matrix and overall accuracy are used to evaluate the performance of the improved LetNet network. To calculate the overall accuracy of the oil product image dataset, the dataset was randomly divided into training and test sets according to a set ratio for evaluation.

[0070] The initial classification types set in the improved LetNet network are 10, which needs to be modified to 4. In addition, the data set is divided into two parts, one as a training set and the other as a validation set. The fluorescence spectra of oil products in the training set need to be preprocessed, while the validation set does not. The initial learning rate is set to 0.02, and the weight parameters of the network are updated and optimized using the Adam algorithm. The batch sizes are set to 24, 64, and 72 respectively, and the image data set needs to be batch normalized according to the batch size. The dropout is set to 0.25, and the epoch is set to 200. During the training process, the network will perform data expansion operations on the oil product images.

[0071] Table 3 shows the effect of batch size on the classification accuracy of the traditional LetNet network, and Table 4 shows the effect of batch size on the classification accuracy of the improved LetNet network. When the learning rate is 0.02 and the batch size is 64, the classification accuracy of the LetNet network is 90.5%, while the classification accuracy of the improved LetNet network reaches 94.0%, which shows that the recognition accuracy of the improved LetNet network is higher than that of the LetNet network.

[0072] Table 3 Effect of batch size on the accuracy of traditional LetNet network classification

[0073]

[0074] Table 4 Effect of batch size on the classification accuracy of the improved LetNet network

[0075]

[0076] For further analysis, this embodiment produces a confusion matrix. In this process, the image dataset is divided into two parts, one as a training set and the other as a validation set, with the division ratios being 50%, 70%, and 90%, respectively. The specific relationship is shown in Table 5.

[0077] Table 5 The impact of training set ratio and batch size on the classification accuracy of the improved LetNet network

[0078]

[0079] When the oil image dataset training accounts for a maximum of 90%, the improved LetNet network classification accuracy is improved overall. When the training image dataset accounts for the same proportion of the overall dataset, the increase in batch size improves the network classification accuracy. Different proportions of training sets have a certain impact on the accuracy of the classification results. When the batch size is 64 and the oil image dataset training accounts for 90%, the improved LetNet network classification accuracy can reach 94%.

[0080] refer to Figure 4 , Figure 4 For the confusion matrix with a training ratio of 90% under the oil image dataset, a total of 1,000 fluorescence spectral image samples were collected, including 250 each of four oil products: lubricating oil, engine oil, transformer oil and motorcycle oil. Only the fluorescence spectral images of 6 oil images were misclassified, 2 fluorescence spectral images of transformer oil were classified as fluorescence spectral images of engine oil, 1 fluorescence spectral image of engine oil was classified as fluorescence spectral image of transformer oil, and 3 fluorescence spectral images of lubricating oil were classified as fluorescence spectral images of engine oil. Due to the activation function layer added in the improved LetNet network, the fitting ability of the network model is enhanced, the extracted feature information is clearer, and the recognition accuracy of the network is further improved.

[0081] Example 2:

[0082] refer to Figure 5 This embodiment provides an oil product identification device based on an improved LetNet network, including the following modules:

[0083] Image acquisition module 1, used to acquire fluorescence spectrum images of various oil products;

[0084] An image preprocessing module 2 is used to preprocess each of the fluorescence spectrum images to obtain preprocessed image data;

[0085] Network improvement module 3 is used to improve the traditional LetNet network to obtain an improved LetNet network;

[0086] A network training module 4 is used to construct a training set based on the preprocessed image data and train the improved LetNet network. After the training is completed, a trained oil product recognition model is obtained;

[0087] The oil product identification module 5 is used to identify the pre-processed fluorescence spectrum image of the oil product to be identified by using the trained oil product identification model, and output the oil product identification classification result.

[0088] Example 3:

[0089] This embodiment provides a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the oil product identification method is executed and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0090] The key technical points of the above embodiments of the present invention are:

[0091] 1. Since there are a lot of high-frequency noise signals in the fluorescence spectrum data collected by the equipment based on the ultraviolet fluorescence method, the present invention uses a low-pass filter to filter the data.

[0092] 2. Compared with the LetNet network results, the improved LetNet network adds multiple layers of activation function layers to the network structure to enhance the network's nonlinear fitting ability.

[0093] 3. During the full connection process, the LetNet network directly connects all neurons, while the improved LetNet network uses dropout to inactivate neurons with a probability of 0.25, and then uses the flatten function, linear function and log_softmax function to complete the network's recognition function. On the one hand, it can reduce the overall parameter amount, and on the other hand, it can effectively prevent overfitting of the network model.

[0094] 4. The convolution kernel size used in the convolution layer of the LetNet network structure is 5*5, while the convolution kernel size used in the convolution layer of the improved LetNet network structure is 3*3, which reduces the overall computational complexity of the network to a certain extent.

[0095] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or system comprising the element.

[0096] The serial numbers of the embodiments of the present invention are for descriptive purposes only and do not represent superiority or inferiority of the embodiments. In a unit claim that lists several means, several of these means may be embodied by the same item of hardware. The use of the terms first, second, and third, etc., does not denote any order and should be construed as identifiers.

[0097] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An oil product identification method based on an improved LetNet network, characterized in that: The following steps are involved: Obtain fluorescence spectrum images of various oil products; Preprocessing each of the fluorescence spectrum images to obtain preprocessed image data; The pre-processing step comprises: Using an FIR filter to filter the fluorescence spectrum image; The pre-processing step further includes: Input the filtered fluorescence spectrum image, first perform channel conversion on the image to remove the transparent channel in the image, then transform the image to change the size of the original image, then randomly rotate the image, and center crop the image. After completing the above steps, convert the image to Tensor image data, then perform normalization and grayscale operation on the image to add a dimension to the Tensor image data, and finally output the preprocessed image data; Improve the traditional LetNet network to obtain an improved LetNet network; The improved LetNet network includes: The sequentially connected input layer, first convolution layer, first activation function layer, second convolution layer, first pooling layer, second activation function layer, third convolution layer, third activation function layer, second pooling layer, fully connected layer and output layer; The first activation function layer, the second activation function layer, and the third activation function layer are used to enhance the nonlinear fitting ability of the network; In the process of full connection of the fully connected layer, dropout is used to inactivate neurons with a probability of 0.25, and the flatten function, linear function and log_softmax function are used to complete the recognition function of the network; Building a training set based on the preprocessed image data, training the improved LetNet network, and obtaining a trained oil product recognition model after the training is completed; The pre-processed fluorescence spectrum image of the oil product to be identified is identified by the trained oil product identification model, and the oil product identification classification result is output.

2. The oil product identification method according to claim 1, characterized in that: The step of obtaining fluorescence spectrum images of a plurality of different oil products comprises: The ultraviolet fluorescence method was used to obtain fluorescence spectrum images of various oil products.

3. The oil product identification method according to claim 1, characterized in that: In the improved LetNet network structure, the convolution kernel size used in the three convolutional layers is 3*3.

4. The oil product identification method according to claim 1, characterized in that: The step of identifying the pre-processed fluorescence spectrum image of the oil product to be identified by using the trained oil product identification model and outputting the oil product identification classification result includes: Preprocess the fluorescence spectrum image of the oil product to be identified; The preprocessed fluorescence spectrum image of the oil product to be identified is input into the trained oil product identification model. After convolution calculation by the first convolution layer, it is fitted by the first activation function layer, and then convolution calculation by the second convolution layer. The output result is subjected to maximum pooling processing by the first pooling layer, fitted by the second activation function layer, convolution calculation by the third convolution layer, fitted by the third activation function layer, and maximum pooling processing by the second pooling layer. Dropout processing is then performed, the flatten function is used to flatten the multidimensional matrix, the Linear function is used for full connection, and finally the log_softmax function is used to complete the identification of the oil product image.

5. The oil product identification method according to claim 1, characterized in that: After the step of outputting the oil product identification and classification results, the following steps are also included: The oil product identification and classification results were evaluated by overall accuracy and confusion matrix.

6. An oil product identification device based on an improved LetNet network, used to implement the oil product identification method according to any one of claims 1 to 5, characterized in that: Includes the following modules: Image acquisition module, used to obtain fluorescence spectrum images of various oil products; An image preprocessing module, used to preprocess each of the fluorescence spectrum images to obtain preprocessed image data; The pre-processing step comprises: Using an FIR filter to filter the fluorescence spectrum image; The pre-processing step further includes: Input the filtered fluorescence spectrum image, first perform channel conversion on the image to remove the transparent channel in the image, then transform the image to change the size of the original image, then randomly rotate the image, and center crop the image. After completing the above steps, convert the image to Tensor image data, then perform normalization and grayscale operation on the image to add a dimension to the Tensor image data, and finally output the preprocessed image data; The network improvement module is used to improve the traditional LetNet network to obtain an improved LetNet network; The improved LetNet network includes: The sequentially connected input layer, first convolution layer, first activation function layer, second convolution layer, first pooling layer, second activation function layer, third convolution layer, third activation function layer, second pooling layer, fully connected layer and output layer; The first activation function layer, the second activation function layer, and the third activation function layer are used to enhance the nonlinear fitting ability of the network; In the process of full connection of the fully connected layer, dropout is used to inactivate neurons with a probability of 0.25, and the flatten function, linear function and log_softmax function are used to complete the recognition function of the network; A network training module is used to construct a training set based on the preprocessed image data and train the improved LetNet network. After the training is completed, a trained oil product recognition model is obtained; The oil product identification module is used to identify the pre-processed fluorescence spectrum image of the oil product to be identified through the trained oil product identification model and output the oil product identification classification result.

7. A computer device, characterized in that: The computer device includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the oil product identification method according to any one of claims 1 to 5 is executed.

Images