A method for evaluating the comfort level of tourists in tourist areas based on a negative entropy convolutional neural network

Abstract

The present invention provides a tourist comfort evaluation method for a tourist area based on a negative entropy convolutional neural network. The method utilizes the ultra-wide-angle shooting characteristics of a drone to obtain more accurate environmental features such as images, sounds, and temperature. A negative entropy calculation method is then used to accurately screen out image samples of tourist areas with potential. The Gabor transform image of the negative entropy area is then combined with stratigraphic information, geological background information, and topographic information of the negative entropy area in a padding matrix filling link of the convolutional neural network to construct a sample matrix of the convolutional neural network, thereby further solidifying the characteristics of the negative entropy area samples. The method finally determines the evaluation value of the tourist comfort of the tourist area in the negative entropy area to be tested by determining the loss function and learning rate of the convolutional neural network and training the convolutional neural network, thereby realizing efficient and accurate evaluation and analysis of massive drone data.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent management, and in particular relates to a method for evaluating the comfort level of tourists in tourist areas based on a negative entropy convolutional neural network. Background Art

[0002] Aerial photography has always been expensive and time-consuming. Applying drone technology to aerial photography of tourist areas or exploring remote locations can effectively reduce costs and shorten filming time. Currently, drone aerial photography technology is gaining increasing attention and is being applied in various fields. Because drones are unmanned, they can operate in complex and harsh environments. This advantage has led to their widespread use in evaluating visitor comfort in tourist areas.

[0003] Entropy is an indicator of the degree of disorder in a system. In most cases, entropy is constantly increasing. Negative entropy refers to the presence of a special factor within a system that enables self-repair and maintains good order. By identifying areas of negative entropy in complex systems, we can uncover this factor within the system that enables self-repair and maintains order. This factor can then be fed back to all areas of increasing entropy within the system, striving to improve the system's internal repair and regulatory capabilities.

[0004] Convolutional neural networks possess powerful feature extraction and compression capabilities. They use two data compression methods: maximum pooling and average pooling. These methods dramatically reduce the size of feature maps to within 25% of their original capacity. For example, if an image of a regional sample is input, its features are continuously enhanced through convolution. Each convolution is followed by activation, and then the pooling layer is used. This is a process of continuously extracting features one by one. The final output feature vector serves as the basis for future classification or evaluation.

[0005] With the popularization of drone technology, deeper exploration of the changing trends in the natural laws, historical evolution, tourism geology, environmental geology and other internal mechanisms of tourist areas has become a new topic currently faced by tourism enthusiasts and geological researchers. The integration of drone aerial photography with convolutional neural networks and negative entropy theory algorithms can be used to evaluate these unknown factors in tourist areas. Summary of the Invention

[0006] The purpose of the present invention is to provide a tourist comfort evaluation method for tourist areas based on negative entropy convolutional neural network, and to achieve efficient and accurate evaluation and analysis of massive data from drones by integrating negative entropy theory with convolutional neural network. In order to achieve the above object, the technical solution of the present invention is: a method for evaluating the comfort level of tourists in tourist areas based on a negative entropy convolutional neural network, comprising the following steps:

[0007] Step S0: Collect UAV aerial images and convert the collected images into grayscale 256 colors;

[0008] Step S1: Calculate the entropy value contained in the image after S0 processing;

[0009] Step S2: Calculate the negative entropy area through four consecutive shots, extract the negative entropy area, and construct a negative entropy area sparse matrix;

[0010] Step S3: Perform Gabor transform on the image in the negative entropy area;

[0011] Step S4: Combine the grayscale image that has undergone saliency processing with the terrain, strata, and geological information of the region in the padding matrix filling link of the convolutional neural network to construct a convolutional neural network sample matrix;

[0012] Step S5: Send a large number of representative regional samples into the convolutional neural network, then construct the convolution kernel of the convolutional neural network, construct the feature pooling compression algorithm of the convolutional neural network, construct the loss function of the convolutional neural network, determine the target convergence error accuracy of the neural network, and finally determine the weight of each neuron of the convolutional neural network by training the convolutional neural network on the sample regional data;

[0013] Step S6: Send the feature matrix of a certain negative entropy region into the convolutional neural network, calculate the tourist comfort level of the region and output the tourist comfort level value of the region to the log file, and end the process.

[0014] The drone aerial images are sampled by a wide-angle camera, the focal length of the wide-angle camera lens ranges from 10 to 24 mm, and the aperture ranges from F4 to F13.

[0015] In the steps S1 and S2,

[0016] The mathematical expression of entropy calculation is:

[0017]

[0018] The entropy value of the drone aerial images in the area is obtained through the entropy calculation formula. The priority of judging the entropy reduction trend is as follows:

[0019] a. The fourth entropy value < the third entropy value < the second entropy value < the first entropy value

[0020] b. The fourth entropy value < the second entropy value < the first entropy value

[0021] c. The fourth entropy value < the first entropy value

[0022] If situation a occurs, the region is defined as the highest priority negative entropy region; if situation b occurs, the region is defined as the medium priority negative entropy region; if situation c occurs, the region is defined as the lowest priority negative entropy region;

[0023] The mathematical expression of the total entropy of the drone aerial photography area is:

[0024]

[0025] When the drone takes aerial photos a certain number of times and forms a continuous time series, the total entropy integral of the time series of the aerial photography area is expressed as:

[0026] .

[0027] The negative entropy convolutional neural network tourist comfort evaluation method meets the conditions for extracting sparse matrices:

[0028]

[0029] k is the ratio of the total number of negative entropy regions to the total number of all regions. When the ratio k is less than or equal to 0.05, the condition for extracting the sparse matrix is triggered.

[0030] In step S3, the image saliency is calculated based on Gabor transform, and the mathematical expression of Gabor transform is:

[0031]

[0032] The real part of the Gabor transform is:

[0033]

[0034]

[0035]

[0036] is the wavelength, For direction, is the aspect ratio, is the standard deviation of the Gaussian kernel function. The saliency map of the regional image is obtained by convolution of the real part of the Gabor transform with the grayscale image. Its mathematical expression is:

[0037]

[0038] in is the original grayscale image, is the real part of Gabor transform, is the saliency map.

[0039] In step S4, in the convolutional neural network sample matrix padding matrix filling link,

[0040] The padding matrix filling link of the convolutional neural network sample matrix contains stratum information, topographic information, geological background information, stratum information The value range is between 0 and 1, terrain information The value range is between 0 and 1, geological background information The value range is between 0 and 1. During the matrix padding process of the convolutional neural network, the padding value Stratigraphic information , terrain information and geological background information The three elements are respectively multiplied by the coefficients of each element and the sum of them. The mathematical expression for calculating the padding matrix filling value is:

[0041] .

[0042] In step S5, the convolutional neural network includes 3 convolutional layers, 2 pooling layers, and 1 fully connected layer. The error back propagation algorithm is used to maximize the verification accuracy, and the momentum is between 0.75 and 0.85.

[0043] In step S5, the feature compression method of the first pooling layer of the convolutional neural network is the maximum pooling algorithm, and the feature compression method of the second pooling layer is the maximum pooling algorithm.

[0044] In step S5, the learning rate optimization mode of the negative entropy convolutional neural network is an exponential decay mode, and the initial learning rate is between 0.55 and 0.65.

[0045] In step S5, the loss function of the negative entropy convolutional neural network is a piecewise loss function. When the difference between the evaluation value and the theoretical value is greater than 0.2, the negative entropy coefficient negentropy is assigned a value of 1.5; when the difference between the evaluation value and the theoretical value is greater than 0 and less than 0.2, the ordinary coefficient is assigned a value of 1.2; when the theoretical value is greater than the evaluation value, the disordered coefficient is assigned a value of 0.8. The mathematical expression of the loss function is:

[0046] .

[0047] The present invention utilizes the ultra-wide-angle shooting characteristics of drones to obtain more accurate environmental characteristics such as images, sounds, and temperature. Then, through the negative entropy calculation method, it accurately screens image samples of potential tourist areas. The Gabor transform image of the negative entropy region is then combined with the stratigraphic information, geological background information, and topographic information of the negative entropy region in the padding matrix filling link of the convolutional neural network to construct the sample matrix of the convolutional neural network. This further solidifies the characteristics of the negative entropy region samples. Then, by determining the loss function and learning rate of the convolutional neural network and training the convolutional neural network, the comfort evaluation value of tourists in the tourist area to be tested is finally determined. This method realizes efficient and accurate evaluation and analysis of massive data from drones, facilitating precise and efficient management by managers. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The present invention constructs a schematic diagram of a negative entropy regional sparse matrix.

[0049] Figure 2 The priority range of the negative entropy region of the present invention.

[0050] Figure 3 Grayscale effect diagram of negative entropy region taken by drone aerial photography of the present invention.

[0051] Figure 4 Gabor transform saliency effect diagram of drone aerial photography according to the present invention.

[0052] Figure 5 The invention shows the filling effect diagram of the negative entropy regional padding matrix by the drone aerial photography.

[0053] Figure 6 Schematic diagram of the sample matrix construction of the convolutional neural network of the present invention.

[0054] Figure 7 Flowchart of the method of the present invention.

[0055] Figure 8 A graph showing the characteristic information measurement standard for the negative entropy convolutional neural network of the present invention.

[0056] Figure 9 A diagram of the hierarchical structure of the convolutional neural network of the present invention. DETAILED DESCRIPTION

[0057] A tourist comfort evaluation method based on negative entropy convolutional neural network, such as Figures 1 to 9 As shown, the following steps are included:

[0058] Step S0: UAV aerial image collection. The UAV uses a wide-angle camera to take four aerial photos of a scenic spot. The focal length of the wide-angle camera lens ranges from 10 to 24 mm, and the aperture range is from F4 to F13. Figure 3 As shown, the size of the image is 866*866, and the collected image is converted to grayscale 256 colors;

[0059] Step S1: Calculate the entropy value contained in the image after S0 processing;

[0060] Step S2: Calculate the negative entropy area through four consecutive shots, extract the negative entropy area, and construct a sparse matrix of the negative entropy area, such as Figure 1 ;

[0061] In steps S1 and S2, the entropy value calculation is mathematically expressed as:

[0062]

[0063] The entropy value of the drone aerial image in the area is obtained through the entropy calculation formula. The priority of judging the entropy value reduction trend is as follows: Figure 2 :

[0064] a. The fourth entropy value < the third entropy value < the second entropy value < the first entropy value

[0065] b. The fourth entropy value < the second entropy value < the first entropy value

[0066] c. The fourth entropy value < the first entropy value

[0067] If the situation a occurs, the area is defined as the highest priority negative entropy area; if the situation b occurs, the area is defined as the medium priority negative entropy area; if the situation c occurs, the area is defined as the lowest priority negative entropy area.

[0068] The mathematical expression of the total entropy of the drone aerial photography area is:

[0069]

[0070] When the drone takes aerial photos a certain number of times and forms a continuous time series, the total entropy integral of the time series of the aerial photography area is expressed as:

[0071] .

[0072] The conditions for extracting sparse matrices for the tourist comfort evaluation method using the negative entropy convolutional neural network are as follows:

[0073]

[0074] k is the ratio of the total number of negative entropy regions to the total number of all regions. When the ratio k is less than or equal to 0.05, the condition for extracting the sparse matrix is triggered.

[0075] Step S3: Perform Gabor transform on the image in the negative entropy area;

[0076] Image saliency calculation is based on Gabor transform, and the mathematical expression of Gabor transform is:

[0077]

[0078] The real part of the Gabor transform is:

[0079]

[0080]

[0081]

[0082] is the wavelength, For direction, is the aspect ratio, is the standard deviation of the Gaussian kernel function. The saliency map of the regional image is obtained by convolution of the real part of the Gabor transform with the grayscale image. Its mathematical expression is:

[0083]

[0084] in is the original grayscale image, is the real part of Gabor transform, is a saliency map, such as Figure 4 As shown in FIG, the size of the saliency effect map is 866*866.

[0085] Step S4: Combine the grayscale image that has undergone saliency processing with the terrain, strata, and geological information of the region in the padding matrix filling link of the convolutional neural network to construct a convolutional neural network sample matrix;

[0086] The construction process of the convolutional neural network sample matrix is as follows: construct the sample matrix padding matrix filling value Measurement standards, such as Figure 8 As shown,

[0087] In the convolutional neural network sample matrix padding matrix filling link,

[0088] The padding matrix filling link of the convolutional neural network sample matrix contains stratum information, topographic information, geological background information, stratum information The value range is between 0 and 1, terrain information The value range is between 0 and 1, geological background information The value range is between 0 and 1. During the matrix padding process of the convolutional neural network, the padding value Stratigraphic information , terrain information and geological background information The three elements are respectively multiplied by the coefficients of each element and the sum of them. The mathematical expression for calculating the padding matrix filling value is:

[0089]

[0090] Through the above method, the padding matrix filling value can be obtained for the region for:

[0091]

[0092] After matrix filling, a matrix with a width of 11 is wrapped around the saliency effect map after Gabor transformation. The value of each element of the wrapped matrix is 0.765, as shown in the following example: Figure 5 As shown, the size of the filled sample matrix is 887*887, and the padding matrix filling diagram is as follows Figure 6 shown.

[0093] Step S5: A large number of representative regional samples are fed into a convolutional neural network. The convolutional neural network's convolution kernel, feature pooling compression algorithm, and loss function are then constructed. The target convergence error accuracy of the neural network is determined. The convolutional neural network is trained on the sample regional data to ultimately determine the weights of each neuron in the convolutional neural network. The convolutional neural network consists of three convolutional layers, two pooling layers, and one fully connected layer. An error backpropagation algorithm is used to maximize verification accuracy, with a momentum between 0.75 and 0.85. The feature compression method for the first pooling layer of the convolutional neural network is the maximum pooling algorithm, and the feature compression method for the second pooling layer is the maximum pooling algorithm.

[0094] The learning rate optimization method of the convolutional neural network is the exponential decay mode, and the initial learning rate range is between 0.55 and 0.65. When training the model, it is usually encountered that a relatively appropriate learning rate is selected after balancing the training speed and loss of the model, but the loss of the training set stops decreasing after it drops to a certain level, and the training loss may appear in a fixed interval. The situation of oscillation, and the training loss cannot drop further. In this case, it can usually be repaired by appropriately lowering the learning rate. However, lowering the learning rate will extend the time required for training. Learning rate decay is a solution that can balance the contradiction between the two. The basic idea of learning rate decay is that the learning rate will gradually decay as the training progresses. The learning rate optimization method of this negative entropy convolutional neural network is the exponential decay learning rate calculation mode, and the exponential decay learning rate calculation formula is as follows:

[0095]

[0096] Where decayed_learning_rate is the learning rate used in each round of optimization, learning_rate is the initial learning rate set in advance, global_step is the neural network calculation round counter, decay_rate is the decay coefficient, and decay_steps is the decay speed.

[0097] The specific process for determining the sliding step size of the convolutional neural network convolution kernel is as follows: If the sliding step size of the convolution kernel is too large, useful feature information of the input information will be lost. If the sliding step size of the convolution kernel is too small, the resulting feature compression matrix will be too large. Therefore, the process of determining the sliding step size of the convolution kernel is a very cautious process.

[0098] The loss function of the convolutional neural network is a segmented loss function, and the loss function of the convolutional neural network is defined as:

[0099]

[0100] In the entire drone navigation area, there are very few samples in the negative entropy area. In the few negative entropy areas, after the constructed input sample matrix is fed into the convolutional neural network, the probability of obtaining high-quality features will be even smaller. Therefore, this negative entropy convolutional neural network customizes different coefficients for different segments and uses a piecewise loss function to train the neural network. The gradient descent direction of the convolutional neural network loss function moves towards the direction with negative entropy features. The piecewise loss function is shown below:

[0101] When the difference between the evaluation value and the theoretical value is greater than 0.2, the negative entropy coefficient negentropy is assigned a value of 1.5; when the difference between the evaluation value and the theoretical value is greater than 0 and less than 0.2, the ordinary coefficient is assigned a value of 1.2; when the theoretical value is greater than the evaluation value, the disordered coefficient is assigned a value of 0.8. The mathematical expression of the loss function is:

[0102]

[0103] Step S6: The feature matrix of the negative entropy region to be tested, which is sized 887*887, is sent to the trained convolutional neural network. After passing through the fully connected layer and the softmax layer, a number between 1 and 10 is finally obtained. The larger the number, the higher the comfort level of tourists in the negative entropy region. The result is output to the log file, and the process ends.

Claims

1. A tourist comfort evaluation method based on negative entropy convolutional neural network, characterized in that: The following steps are involved: Step S0: Collect UAV aerial images and convert the collected images into grayscale 256 colors; Step S1: Calculate the entropy value contained in the image after S0 processing; Step S2: Calculate the negative entropy area through four consecutive shots, extract the negative entropy area, and construct a negative entropy area sparse matrix; Step S3: Perform Gabor transform on the image in the negative entropy area; Step S4: Combine the grayscale image after saliency processing with the terrain, stratum, and geological information of the area in the padding matrix filling link of the convolutional neural network to construct the convolutional neural network sample matrix. Specifically, the padding matrix filling link of the convolutional neural network sample matrix includes stratum information, topographic information, geological background information, stratum information The value range is between 0 and 1, terrain information The value range is between 0 and 1, geological background information The value range is between 0 and 1. During the matrix padding process of the convolutional neural network, the padding value From stratigraphic information , terrain information and geological background information The three elements are respectively multiplied by the coefficients of each element and the sum of them. The mathematical expression for calculating the padding matrix filling value is: ； Step S5: Send a large number of representative regional samples into the convolutional neural network, then construct the convolution kernel of the convolutional neural network, construct the feature pooling compression algorithm of the convolutional neural network, construct the loss function of the convolutional neural network, determine the target convergence error accuracy of the neural network, and finally determine the weight of each neuron of the convolutional neural network by training the convolutional neural network on the sample regional data; Step S6: Send the feature matrix of a certain negative entropy region into the convolutional neural network, calculate the tourist comfort level in the region, and output the tourist comfort level value in the region to the log file, ending the process.

2. The method for evaluating tourist comfort in tourist areas based on negative entropy convolutional neural network according to claim 1 is characterized in that: The drone aerial images are sampled by a wide-angle camera, the focal length of the wide-angle camera lens ranges from 10 to 24 mm, and the aperture ranges from F4 to F13.

3. The tourist comfort evaluation method for tourist areas based on negative entropy convolutional neural network according to claim 1 is characterized in that: In the steps S1 and S2, The mathematical expression of entropy calculation is: ， The entropy value of the drone aerial images in the area is obtained through the entropy calculation formula. The priority of judging the entropy reduction trend is as follows: a. The fourth entropy value < the third entropy value < the second entropy value < the first entropy value b. The fourth entropy value < the second entropy value < the first entropy value c. The fourth entropy value is less than the first entropy value If situation a occurs, the area is defined as the highest priority negative entropy area; if situation b occurs, the area is defined as the medium priority negative entropy area; if situation c occurs, the area is defined as the lowest level negative entropy area.

4. The method for evaluating tourist comfort in tourist areas based on negative entropy convolutional neural network according to claim 1 is characterized in that: The negative entropy convolutional neural network tourist comfort evaluation method meets the conditions for extracting sparse matrices: ， k is the ratio of the total number of negative entropy regions to the total number of all regions. When the ratio k is less than or equal to 0.05, the condition for extracting the sparse matrix is triggered.

5. The method for evaluating tourist comfort in tourist areas based on negative entropy convolutional neural network according to claim 1 is characterized in that: In step S3, the image saliency is calculated based on Gabor transform, and the mathematical expression of Gabor transform is: ， The real part of the Gabor transform is: ， is the wavelength, For direction, is the aspect ratio, is the standard deviation of the Gaussian kernel function. The saliency map of the regional image is obtained by convolution of the real part of the Gabor transform with the grayscale image. Its mathematical expression is: ， in is the original grayscale image, is the real part of Gabor transform, is the saliency map.

6. The method for evaluating tourist comfort in tourist areas based on negative entropy convolutional neural network according to claim 1 is characterized in that: In step S5, the convolutional neural network includes 3 convolutional layers, 2 pooling layers, and 1 fully connected layer. The error back propagation algorithm is used to maximize the verification accuracy, and the momentum is between 0.75 and 0.

85.

7. The method for evaluating tourist comfort in tourist areas based on negative entropy convolutional neural network according to claim 1 is characterized in that: In step S5, the feature compression method of the first pooling layer of the convolutional neural network is the maximum pooling algorithm, and the feature compression method of the second pooling layer is the maximum pooling algorithm.

8. The method for evaluating tourist comfort in tourist areas based on negative entropy convolutional neural network according to claim 1 is characterized in that: In step S5, the learning rate optimization mode of the negative entropy convolutional neural network is an exponential decay mode, and the initial learning rate is between 0.55 and 0.

65.

9. The method for evaluating tourist comfort in tourist areas based on negative entropy convolutional neural network according to claim 1 is characterized in that: In step S5, the loss function of the negative entropy convolutional neural network is a piecewise loss function. When the difference between the evaluation value and the theoretical value is greater than 0.2, the negative entropy coefficient negentropy is assigned a value of 1.5; when the difference between the evaluation value and the theoretical value is greater than 0 and less than 0.2, the ordinary coefficient is assigned a value of 1.2; when the theoretical value is greater than the evaluation value, the disordered coefficient is assigned a value of 0.

8. The mathematical expression of the loss function is: 。

Images