Airport runway foreign matter detection and analysis method based on artificial intelligence

By introducing convolutional layer and attention mechanism into the YOLO network and combining data enhancement technology, the problem of missed detection of foreign matter detection on airport runways in harsh environments has been solved, and the detection accuracy and generalization ability have been improved.

CN118982811APending Publication Date: 2024-11-19SHANDONG EAGLE INFORMATION ENG CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411073563.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

The existing airport runway foreign object detection methods are likely to lead to missed and misdetection during night or in severe weather conditions, especially for small foreign object detection.

Method used

Using an artificial intelligence-based method, the introduction of convolutional layer and attention mechanisms through improved YOLO network, combined with data enhancement technology, a robust and generalized model is built for foreign body detection. Specific steps include shooting and labeling image data sets, performing data augmentation processing, dividing training and test sets, and introducing improved convolutional layers and attention mechanisms on the basis of traditional YOLO networks for training and deployment.

Benefits of technology

It significantly improves the detection accuracy of the model and the recognition ability of small foreign objects, enhances the generalization ability under different environmental conditions, and reduces the occurrence of missed detection and missed detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982811A_ABST
    Figure CN118982811A_ABST
Patent Text Reader

Abstract

The invention relates to the field, in particular to an airport runway foreign matter detection and analysis method based on artificial intelligence. The method comprises the following steps: S1, shooting and collecting an airport runway image, constructing an image data set, and labeling foreign matters by using Labe l Img; s2, carrying out image overturning, blurring and brightness change enhancement processing on the marked image data set; s3, randomly dividing the enhanced image data set into a training set and a test set according to a proportion of 8: 2; and S4, introducing an improved convolutional layer and an attention mechanism on the basis of a traditional YOLO network structure, training and optimizing the network by using the training set, and deploying the network to an actual environment for foreign matter detection. According to the airport runway foreign matter detection and analysis method based on artificial intelligence, an improved convolutional layer and an attention mechanism are introduced into a traditional YOLO network, and a training set is used for training and optimization, so that the detection precision and the recognition capability of small foreign matters are remarkably improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an airport runway foreign body detection and analysis method based on artificial intelligence. Background Art

[0002] An artificial intelligence-based airport runway foreign object detection and analysis method aims to achieve efficient and accurate detection of runway foreign objects by introducing convolutional layers and attention mechanisms through an improved YOLO network, initialize the model with weights pre-trained with large-scale datasets, and combine data augmentation technology to improve the robustness and generalization ability of the model.

[0003] Existing airport runway foreign object detection and analysis methods include manual inspection and radar detection technology. However, when monitoring and detecting small foreign objects at night or in bad weather conditions, due to the changing environmental conditions and the small size of foreign objects, it is easy to cause missed detection and false detection. Therefore, an airport runway foreign object detection and analysis method based on artificial intelligence is provided. Summary of the invention

[0004] The purpose of the present invention is to provide an airport runway foreign object detection and analysis method based on artificial intelligence to solve the problem of missed detection and false detection due to the ever-changing environmental conditions and the small size of foreign objects mentioned in the above background technology.

[0005] To achieve the above object, the present invention provides an airport runway foreign body detection and analysis method based on artificial intelligence, comprising the following steps:

[0006] S1. Take and collect airport runway images, construct image datasets and use LabelImg to label foreign objects;

[0007] S2, performing image flipping, blurring, and brightness change enhancement processing on the labeled image dataset;

[0008] S3, randomly divide the enhanced image dataset into training set and test set in a ratio of 8:2;

[0009] S4. Based on the traditional YOLO network structure, an improved convolutional layer and attention mechanism are introduced, the network is trained and optimized using the training set, and then deployed in the actual environment for foreign object detection.

[0010] As a further improvement of the present technical solution, the specific steps involved in constructing the image data set in S1 are as follows:

[0011] S1.1. Use a telephoto camera to capture images of foreign objects on the airport runway, including close-up, long-distance, clear, and blurred images:

[0012] S1.2. Preprocess the airport runway foreign body images, including image denoising, image normalization and resolution adjustment, and construct the image dataset D original ;

[0013] S1.3. Use the Label Img tool to manually label the preprocessed foreign body images;

[0014] S1.4. Image dataset D original and the corresponding annotation files are organized and stored in folders.

[0015] As a further improvement of the technical solution, the specific steps involved in the enhancement processing of the labeled image data set in S2 are as follows:

[0016] S2.1. For the labeled original image dataset D original Perform flipping, blurring, and brightness change enhancement processing to obtain an enhanced image dataset D aunmented ;

[0017] S2.2. The original image dataset D original With the enhanced image dataset D augmented Integrate into a dataset D final , used for subsequent training and testing.

[0018] As a further improvement of the technical solution, the specific steps of constructing the training set and the test set in S3 are as follows:

[0019] S3.1. Determine the total number of samples in the data set as N, and calculate the number of training set samples N according to the ratio of 8:2 train And the number of test set samples N test ;

[0020] S3.2, use the Fisher-Yates shuffle algorithm to shuffle the data set D final Random shuffle;

[0021] S3.3. Select the top N from the randomly shuffled data set train samples as training set D train , the remaining N test samples as the test set D test .

[0022] As a further improvement of this technical solution, the number of samples involved in S3.1 is as follows:

[0023] Total number of samples N:

[0024] N=|D final |;

[0025] The number of training set samples Ntrain :

[0026] N train =[0.8×N];

[0027] The number of test set samples N test :

[0028] N test =NN train .

[0029] As a further improvement of the technical solution, the Fisher-Yates shuffling algorithm is used in S3.2 to shuffle the data set D final The specific steps and mathematical formulas involved in random shuffling are as follows:

[0030] S3.2.1. Initialize the array:

[0031]

[0032] S3.2.2. Generate a random integer j:

[0033] j = [(i+1)·rand()];

[0034] S3.2.3. Exchange elements:

[0035]

[0036] As a further improvement of the technical solution, the steps of using the improved YOLO network for training in 4 are as follows:

[0037] S4.1. Based on the traditional YOLO network structure, an improved convolutional layer and attention mechanism are introduced;

[0038] S4.2. Through transfer learning, the weights of the YOLO model pre-trained on a large dataset are loaded into the improved YOLO network.

[0039] S4.3. Use the AdamW optimizer and set the appropriate learning rate, batch size, and number of training rounds.

[0040] S4.4. Use the training set D divided in S3 train , and introduce the attention mechanism to train the improved YOLO network, using the output of the single-head attention mechanism and the output of the multi-head attention mechanism as the input of the multi-layer perceptron:

[0041] S4.5. Use the test set D divided in S3 test Evaluate the trained model using the following evaluation metrics: accuracy, recall, precision, and F1 score.

[0042] S4.6. Adjust and optimize the model structure, hyperparameters and training methods based on the evaluation results to improve detection performance;

[0043] S4.7. Deploy the trained improved YOLO network into the actual environment for the task of detecting foreign objects on airport runways.

[0044] S4.8. Monitor model performance in real-world scenarios and make adjustments and optimizations as needed.

[0045] As a further improvement of the technical solution, the mathematical model formula involved in the improved convolutional layer introduced in S4.1 is as follows:

[0046] Depthwise Separable Convolution:

[0047] Conv standard (x)≈Conv depthwise (x)+Conv pointwise (x);

[0048] Among them, Conv standard (x) is the standard convolution; Conv depthwise (x) is the depth convolution; Conv pointwise (x) is point-wise convolution;

[0049] Residual Connection:

[0050] y=F(x,{W i})+x;

[0051] Among them, F(x, {W i}) is the mapping of the convolutional layer; x is the input; y is the output;

[0052] The mathematical model formula involved in the attention mechanism introduced in S4.1 is as follows:

[0053] Single-head attention mechanism:

[0054]

[0055] Where Q is the query matrix; K is the key matrix; V is the value matrix; d K is the dimension of the key vector;

[0056] Multi-head attention mechanism:

[0057] MultiHead(Q,K,V)=Concat(head 1 , head 2 , ..., head h )W O ;

[0058] Among them, headi =Attention(QW i Q , KW i K , VW i V ), W i Q , W i K , W i K and W O is a learnable weight matrix.

[0059] As a further improvement of this technical solution, the training set D is used in S4.4 train The specific steps involved in training the modified YOLO network are as follows:

[0060] S4.4.1. Input training set D train After the convolution layer, improved convolution layer and attention mechanism, the feature map is obtained:

[0061] S4.4.2. Calculate the loss between the predicted result and the true annotation, including positioning error, classification error and confidence error:

[0062] L=L loc +L cls +L conf ;

[0063] S4.4.3. Calculate the gradient through the back-propagation algorithm and update the network weights;

[0064] S4.4.4. Use AdamW optimizer to update weights:

[0065]

[0066] S4.4.5. Repeat the process of forward propagation, loss calculation, backpropagation and weight update until the set number of training rounds and loss function convergence are achieved.

[0067] As a further improvement of the technical solution, the test set D is used in S4.5 test The trained model is evaluated for accuracy, recall, precision, and F1 score. The calculation of the indicators involved is as follows:

[0068] Accuracy:

[0069]

[0070] Among them, TP represents the number of samples that are actually positive and predicted to be positive; TN represents the number of samples that are actually negative and predicted to be negative; TN represents the number of samples that are actually negative and predicted to be negative; FN represents the number of samples that are actually positive but predicted to be negative;

[0071] Recall:

[0072]

[0073] F1 score:

[0074]

[0075] Compared with the prior art, the present invention has the following beneficial effects:

[0076] 1. In this method of detecting and analyzing foreign objects on airport runways based on artificial intelligence, by introducing improved convolutional layers and attention mechanisms based on the traditional YOLO network structure, and using training sets to train and optimize the network, the detection accuracy of the model and its ability to identify small foreign objects can be significantly improved.

[0077] 2. In this artificial intelligence-based airport runway foreign object detection and analysis method, the use of data augmentation technology and pre-training weight initialization can increase the diversity of training data, accelerate model convergence, and improve the generalization ability of the model under different environmental conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 The figure is a flow chart of the overall method of the present invention. DETAILED DESCRIPTION

[0079] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0080] Example:

[0081] See also Figure 1 As shown, this embodiment provides an airport runway foreign body detection and analysis method based on artificial intelligence, comprising the following steps:

[0082] S1, capture and collect airport runway images, construct image datasets and use Label Img to mark foreign objects;

[0083] The specific steps involved in constructing the image dataset in S1 are as follows:

[0084] S1.1. Use a telephoto camera to capture images of foreign objects on the airport runway, including close-up, long-distance, clear and blurred images;

[0085] S1.2. Preprocess the airport runway foreign body images, including image denoising, image normalization and resolution adjustment, and construct the image dataset D original ;

[0086] S1.3. Use the Label Img tool to manually label the preprocessed foreign body images;

[0087] S1.4. Image dataset D original and the corresponding annotation files are organized and stored in folders.

[0088] S2, performing image flipping, blurring, and brightness change enhancement processing on the labeled image dataset;

[0089] The specific steps involved in performing enhancement processing on the labeled image dataset in S2 are as follows:

[0090] S2.1. For the labeled original image dataset D original Perform flipping, blurring, and brightness change enhancement processing to obtain an enhanced image dataset D augmented ;

[0091] S2.2. The original image dataset D original With the enhanced image dataset D augmented Integrate into a dataset D final , used for subsequent training and testing.

[0092] S3, randomly divide the enhanced image dataset into training set and test set in a ratio of 8:2;

[0093] The specific steps of constructing the training set and the test set in S3 are as follows:

[0094] S3.1. Determine the total number of samples in the data set as N, and calculate the number of training set samples N according to the ratio of 8:2 train And the number of test set samples N test ;

[0095] S3.2, use the Fisher-Yates shuffle algorithm to shuffle the data set D final Random shuffle;

[0096] S3.3. Select the top N from the randomly shuffled data set train samples as training set D train , the remaining N test samples as the test set D test .

[0097] The number of samples involved in S3.1 is as follows:

[0098] Total number of samples N:

[0099] N=|D final |;

[0100] The number of training set samples N train :

[0101] N train =[0.8×N];

[0102] The number of test set samples N test :

[0103] N test =NN train .

[0104] In S3.2, the Fisher-Yates shuffle algorithm is used to shuffle the dataset D final The specific steps and mathematical formulas involved in random shuffling are as follows:

[0105] S3.2.1. Initialize the array:

[0106]

[0107] S3.2.2. Generate a random integer j:

[0108] j = [(i+1)·rand()];

[0109] S3.2.3. Exchange elements:

[0110]

[0111] S4. Based on the traditional YOLO network structure, an improved convolutional layer and attention mechanism are introduced, the network is trained and optimized using the training set, and then deployed in the actual environment for foreign object detection;

[0112] The traditional YOLO network structure includes an input layer, a convolutional layer, a downsampling layer, a fully connected layer, and an output layer. The key feature is that a single forward propagation can simultaneously predict multiple bounding boxes and category probabilities, and is real-time.

[0113] As described above, an improved convolutional layer and attention mechanism are introduced based on the traditional YOLO network structure. Through deep separable convolution and residual connection, the improved convolutional layer can extract detailed features in the image more efficiently, especially with better robustness to small targets and complex backgrounds; the attention mechanism enables the model to automatically focus on key areas when processing images, thereby improving the accuracy of target detection. The single-head attention mechanism simplifies the attention calculation, and the multi-head attention mechanism further improves the recognition ability of the model by calculating different feature subspaces in parallel; by introducing pre-trained weights and improved network structure, the convergence of the model can be accelerated, the training efficiency can be improved, and the generalization ability of the model can be improved through data enhancement and transfer learning;

[0114] The specific steps of using the improved YOLO network for training in 4 are as follows:

[0115] S4.1. Based on the traditional YOLO network structure, an improved convolutional layer and attention mechanism are introduced;

[0116] S4.2. Through transfer learning, the weights of the YOLO model pre-trained on a large dataset are loaded into the improved YOLO network.

[0117] S4.3. Use the AdamW optimizer and set the appropriate learning rate, batch size, and number of training rounds.

[0118] S4.4. Use the training set D divided in S3 train , and introduce the attention mechanism to train the improved YOLO network, using the output of the single-head attention mechanism and the output of the multi-head attention mechanism as the input of the multi-layer perceptron;

[0119] S4.5. Use the test set D divided in S3 test Evaluate the trained model using the following evaluation metrics: accuracy, recall, precision, and F1 score.

[0120] S4.6. Adjust and optimize the model structure, hyperparameters and training methods based on the evaluation results to improve detection performance;

[0121] S4.7. Deploy the trained improved YOLO network into the actual environment for the task of detecting foreign objects on airport runways.

[0122] S4.8. Monitor model performance in real-world scenarios and make adjustments and optimizations as needed.

[0123] The mathematical model formula involved in the improved convolutional layer introduced in S4.1 is as follows:

[0124] Depthwise Separable Convolution:

[0125] Convstandard (x)≈Conv depthwise (x)+Conv pointwise (x);

[0126] Among them, Conv standard (x) is the standard convolution; Conv depthwise (x) is the depth convolution; Conv pointwise (x) is point-wise convolution;

[0127] Residual Connection:

[0128] y=F(x,{W i})+x;

[0129] Among them, F(x, {W i}) is the mapping of the convolutional layer; x is the input; y is the output;

[0130] The mathematical model formula involved in the attention mechanism introduced in S4.1 is as follows:

[0131] Single-head attention mechanism:

[0132]

[0133] Where Q is the query matrix; K is the key matrix; V is the value matrix; d K is the dimension of the key vector;

[0134] Multi-head attention mechanism:

[0135] MultiHead(Q,K,V)=Concat(head 1 , head 2 , ..., head h )W O ;

[0136] Among them, head i =Attention(QW i Q , KW i K , VW i V ), W i Q , W i K , W i K and W O is a learnable weight matrix.

[0137] The training set D is used in S4.4 train The specific steps involved in training the modified YOLO network are as follows:

[0138] S4.4.1. Input training set D train After the convolution layer, improved convolution layer and attention mechanism, the feature map is obtained;

[0139] S4.4.2. Calculate the loss between the predicted result and the true annotation, including positioning error, classification error and confidence error:

[0140] L=L loc +L cls +L conf ;

[0141] S4.4.3. Calculate the gradient through the back-propagation algorithm and update the network weights;

[0142] S4.4.4. Use AdamW optimizer to update weights:

[0143]

[0144] S4.4.5. Repeat the process of forward propagation, loss calculation, back propagation and weight update until the set number of training rounds and loss function convergence are achieved;

[0145] Among them, the feature map in S4.4.1 is the input training set D train The intermediate representation of data after it has been processed by the convolutional layers in a neural network;

[0146] Wherein, the back propagation algorithm in S4.4.3 is used to calculate the gradient of the loss function relative to the network weights, and update the weights through these gradients to minimize the loss function;

[0147] The test set D is used in S4.5 test The trained model is evaluated for accuracy, recall, precision, and F1 score. The calculation of the indicators involved is as follows:

[0148] Accuracy:

[0149]

[0150] Among them, TP represents the number of samples that are actually positive and predicted to be positive; TN represents the number of samples that are actually negative and predicted to be negative; TN represents the number of samples that are actually negative and predicted to be negative; FN represents the number of samples that are actually positive but predicted to be negative;

[0151] Recall:

[0152]

[0153] F1 score:

[0154]

[0155] Among them, the accuracy rate represents the proportion of samples predicted correctly by the model to the total number of samples; the recall rate represents the proportion of samples that are correctly predicted to be positive among all truly positive samples; the precision rate represents the proportion of samples that are truly positive among all samples predicted to be positive; the F1 score is the harmonic mean of the precision rate and the recall rate, which combines the effects of both.

[0156] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An airport runway foreign body detection and analysis method based on artificial intelligence, characterized by: The following steps are involved: S1. Take and collect airport runway images, construct image datasets and use LabelImg to label foreign objects; S2, performing image flipping, blurring, and brightness change enhancement processing on the labeled image dataset; S3, randomly divide the enhanced image dataset into training set and test set in a ratio of 8:2; S4. Based on the traditional YOLO network structure, an improved convolutional layer and attention mechanism are introduced, the network is trained and optimized using the training set, and then deployed in the actual environment for foreign object detection.

2. The method for detecting and analyzing foreign objects on airport runways based on artificial intelligence according to claim 1 is characterized in that: The specific steps involved in constructing the image dataset in S1 are as follows: S1.

1. Use a telephoto camera to capture images of foreign objects on the airport runway, including close-up, long-distance, clear and blurred images; S1.

2. Preprocess the airport runway foreign body images, including image denoising, image normalization and resolution adjustment, and construct the image dataset D original ; S1.

3. Use the LabelImg tool to manually label the preprocessed foreign body images; S1.

4. Image dataset D original and the corresponding annotation files are organized and stored in folders.

3. The method for detecting and analyzing foreign objects on airport runways based on artificial intelligence according to claim 1 is characterized in that: The specific steps involved in performing enhancement processing on the labeled image dataset in S2 are as follows: S2.

1. For the labeled original image dataset D original Perform flipping, blurring, and brightness change enhancement processing to obtain an enhanced image dataset D augmented ; S2.

2. The original image dataset D original With the enhanced image dataset D augmented Integrate into a dataset D final , used for subsequent training and testing.

4. The method for detecting and analyzing foreign objects on airport runways based on artificial intelligence according to claim 1 is characterized in that: The specific steps of constructing the training set and the test set in S3 are as follows: S3.

1. Determine the total number of samples in the data set as N, and calculate the number of training set samples N according to the ratio of 8:2 train And the number of test set samples N test ; S3.2, use the Fisher-Yates shuffle algorithm to shuffle the data set D final Random shuffle; S3.

3. Select the top N from the randomly shuffled data set train samples as training set D train , the remaining N test samples as the test set D test .

5. The method for detecting and analyzing foreign objects on airport runways based on artificial intelligence according to claim 4 is characterized in that: The number of samples involved in S3.1 is as follows: Total number of samples N: N=|D final |; The number of training set samples N train : The number of test set samples N test : N test =N-N train 。 6. The method for detecting and analyzing foreign objects on airport runways based on artificial intelligence according to claim 4 is characterized in that: In S3.2, the Fisher-Yates shuffle algorithm is used to shuffle the dataset D final The specific steps and mathematical formulas involved in random shuffling are as follows: S3.2.

1. Initialize the array: S3.2.

2. Generate a random integer j: S3.2.

3. Exchange elements:

7. The method for detecting and analyzing foreign objects on airport runways based on artificial intelligence according to claim 1 is characterized by: The specific steps of using the improved YOLO network for training in 4 are as follows: S4.

1. Based on the traditional YOLO network structure, an improved convolutional layer and attention mechanism are introduced; S4.

2. Through transfer learning, the weights of the YOLO model pre-trained on a large dataset are loaded into the improved YOLO network. S4.

3. Use the AdamW optimizer and set the appropriate learning rate, batch size, and number of training rounds. S4.

4. Use the training set D divided in S3 train , and introduce the attention mechanism to train the improved YOLO network, using the output of the single-head attention mechanism and the output of the multi-head attention mechanism as the input of the multi-layer perceptron; S4.

5. Use the test set D divided in S3 test Evaluate the trained model using the following evaluation metrics: accuracy, recall, precision, and F1 score. S4.

6. Adjust and optimize the model structure, hyperparameters and training methods based on the evaluation results to improve detection performance; S4.

7. Deploy the trained improved YOLO network into the actual environment for the task of detecting foreign objects on airport runways. S4.

8. Monitor model performance in real-world scenarios and make adjustments and optimizations as needed.

8. The method for detecting and analyzing foreign objects on airport runways based on artificial intelligence according to claim 7 is characterized in that: The mathematical model formula involved in the improved convolutional layer introduced in S4.1 is as follows: Depthwise Separable Convolution: Conv standard (x)≈Conv depthwise (x)+Conv pointwise (x); Among them, Conv standard (x) is the standard convolution; Conv depthwise (x) is the depth convolution; Conv pointwise (x) is point-wise convolution; Residual Connection: y=F(x,{W i })+x; Among them, F(x,{W i }) is the mapping of the convolutional layer; x is the input; y is the output; The mathematical model formula involved in the attention mechanism introduced in S4.1 is as follows: Single-head attention mechanism: Where Q is the query matrix; K is the key matrix; V is the value matrix; d K is the dimension of the key vector; Multi-head attention mechanism: MultiHead(Q,K,V)=Concat(head1,head2,…,head h )W O ; in, and W O is a learnable weight matrix.

9. The method for detecting and analyzing foreign objects on airport runways based on artificial intelligence according to claim 7 is characterized in that: The training set D is used in S4.4 train The specific steps involved in training the modified YOLO network are as follows: S4.4.

1. Input training set D train After the convolution layer, improved convolution layer and attention mechanism, the feature map is obtained; S4.4.

2. Calculate the loss between the predicted result and the true annotation, including positioning error, classification error and confidence error: L=L loc +L cls +L conf ; S4.4.

3. Calculate the gradient through the back-propagation algorithm and update the network weights; S4.4.

4. Use AdamW optimizer to update weights: S4.4.

5. Repeat the process of forward propagation, loss calculation, backpropagation and weight update to achieve the set number of training rounds and loss function convergence.

10. The method for detecting and analyzing foreign objects on airport runways based on artificial intelligence according to claim 7, characterized in that: The test set D is used in S4.5 test The trained model is evaluated for accuracy, recall, precision, and F1 score. The calculation of the indicators involved is as follows: Accuracy: Among them, TP represents the number of samples that are actually positive and predicted to be positive; TN represents the number of samples that are actually negative and predicted to be negative; TN represents the number of samples that are actually negative and predicted to be negative; FN represents the number of samples that are actually positive but predicted to be negative; Recall: F1 score: