Infrared ship identification method, system and device and storage medium

Abstract

The invention discloses an infrared ship identification method, system and device, and a storage medium. The method comprises the steps of obtaining a to-be-identified infrared ship image; inputting an infrared ship image to be recognized into the trained student model to obtain a recognition result of the infrared ship image; the trained student model is obtained by the following steps: obtaining a clean infrared ship sample; obtaining a random noise vector group; acquiring an enhanced infrared ship sample according to the clean infrared ship sample and the random noise vector group; and training a student model according to the clean infrared ship sample, the enhanced infrared ship sample and the teacher model subjected to confrontation training to obtain a student model subjected to training. According to the invention, the accuracy of infrared ship image recognition can be improved.

Description

Technical Field

[0001] The present application relates to the field of underwater observation technology, and in particular to an infrared ship recognition method, system, device and storage medium. Background Art

[0002] At present, although infrared imaging technology can effectively reduce the impact of weather and lighting conditions, in actual applications, infrared ship images are still subject to interference from natural phenomena such as sea surface reflection and cloud obstruction, and there is a problem of insufficient recognition accuracy in the process of infrared ship image recognition. Summary of the invention

[0003] The present application aims to propose an infrared ship recognition method, system, device and storage medium, which can improve the accuracy of infrared ship image recognition.

[0004] According to the first aspect of the present application, the infrared ship recognition method includes: Acquire infrared ship images to be identified; Inputting the infrared ship image to be identified into the trained student model to obtain a recognition result of the infrared ship image; The student model that has completed training is obtained by the following steps: Acquire a clean infrared ship sample, wherein the clean infrared ship sample includes a plurality of identifiable clean infrared ship images; Acquire a random noise vector group, wherein the random noise vector group includes a plurality of random noise vectors; Acquire enhanced infrared ship samples according to the clean infrared ship samples and the random noise vector group, wherein the enhanced infrared ship samples have the same characteristic distribution as the clean infrared ship samples; The student model is trained according to the clean infrared ship samples, the enhanced infrared ship samples and the teacher model that has completed adversarial training to obtain the trained student model.

[0005] According to some embodiments of the present application, the step of training a student model based on the clean infrared ship samples, the enhanced infrared ship samples, and a teacher model that has completed adversarial training to obtain the trained student model includes: Inputting the training samples into the teacher model that has completed adversarial training to obtain a first output distribution result, wherein the training samples include the clean infrared ship samples and the enhanced infrared ship samples, and the teacher model that has completed adversarial training is obtained by inputting adversarial samples into the teacher model to complete adversarial training; Input the training samples into the student model to obtain the second output distribution result; Inputting the training sample and the adversarial sample into the student model to obtain a third output distribution result; The student model is trained according to the first output distribution result, the second output distribution result and the third output distribution result to obtain the trained student model.

[0006] According to some embodiments of the present application, obtaining enhanced infrared ship samples according to the clean infrared ship samples and the random noise vector group includes: Inputting the i-th random noise vector in the random noise vector group and the clean infrared ship sample into the generator model to obtain an intermediate infrared ship sample corresponding to the i-th random noise vector, wherein the intermediate infrared ship sample includes a plurality of intermediate infrared ship images, and i is a positive integer; Determine a first loss situation corresponding to the i-th random noise vector according to the intermediate infrared ship samples and the clean infrared ship samples corresponding to the i-th random noise vector; In the case where the first loss situation does not satisfy the first training stop condition of the generator model, adjusting the parameters of the generator model to obtain the updated generator model, and updating i to i+1, returning to the step of inputting the i-th random noise vector in the random noise vector group and the clean infrared ship sample into the generator model until the first training stop condition is satisfied, thereby obtaining the generator model that has completed training; Acquire an optimal noise vector, where the optimal noise vector is one of the random noise vectors in the random noise vector group; The optimal noise vector and the clean infrared ship sample are input into the generator model that has completed training to obtain the enhanced infrared ship sample.

[0007] According to some embodiments of the present application, the optimal noise vector is constrained by the following mathematical expression: ; in, is the optimal noise vector, Z is the random noise vector group, is the generator model that has completed training, and x is the clean infrared ship sample.

[0008] According to some embodiments of the present application, training the student model according to the first output distribution result, the second output distribution result, and the third output distribution result includes: Obtaining a second loss situation according to the first output distribution result, the second output distribution result, and the third output distribution result; When the second loss situation does not meet the second training stop condition of the student model, adjust the parameters of the student model to obtain the updated student model, return to the first output distribution result, the second output distribution result and the third output distribution result to obtain the second loss situation until the second training stop condition is met to obtain the student model that has completed training.

[0009] According to some embodiments of the present application, the second loss condition is constrained by the following mathematical expression: ; Where S is the student model, T is the teacher model that has completed adversarial training, is an adjustable parameter between 0 and 1. Indicates the first output distribution result and the second output distribution result Divergence, Indicates the second output distribution result and the third output distribution result Divergence.

[0010] According to some embodiments of the present application, the adversarial sample is constrained by the following mathematical expression: ; in, is the adversarial sample, x is the clean infrared ship sample, is the preset disturbance value, is a constraint condition, representing an adversarial sample With the original sample The difference between The preset disturbance value is not exceeded. , f is a parameter The adversarial generation sample model, It represents maximizing the output error of the adversarial sample after passing through the model, and y is the true label of the clean infrared ship sample.

[0011] According to the infrared ship identification system of the second aspect of the present application, the system includes: An acquisition module, used for acquiring infrared ship images to be identified; A recognition module, used for inputting the infrared ship image to be recognized into the trained student model to obtain a recognition result of the infrared ship image; The student model that has completed training is obtained by the following steps: Acquire a clean infrared ship sample, wherein the clean infrared ship sample includes a plurality of identifiable clean infrared ship images; Acquire a random noise vector group, wherein the random noise vector group includes a plurality of random noise vectors; Acquire enhanced infrared ship samples according to the clean infrared ship samples and the random noise vector group, wherein the enhanced infrared ship samples have the same characteristic distribution as the clean infrared ship samples; The student model is trained according to the clean infrared ship samples, the enhanced infrared ship samples and the teacher model that has completed adversarial training to obtain the trained student model.

[0012] An electronic device according to an embodiment of the third aspect of the present application includes a processor and a memory, wherein the memory stores programs or instructions that can be run on the processor, and when the programs or instructions are executed by the processor, the steps of the infrared ship identification method as described in the embodiment of the first aspect are implemented.

[0013] According to the computer-readable storage medium of the fourth aspect embodiment of the present application, computer-executable instructions are stored, and the computer-executable instructions are used to execute the infrared ship recognition method as described in the first aspect embodiment above.

[0014] In an embodiment of the present application, a student model is trained to obtain a trained student model suitable for infrared ship recognition. This model is used for recognition to effectively improve the accuracy of infrared ship image recognition. In the training process of the student model, clean infrared ship samples, enhanced infrared ship samples generated from clean infrared ship samples, and a teacher model that has completed adversarial training are used to enhance the diversity of infrared ship images, improve the robustness and generalization ability of the model, and thereby improve the accuracy of infrared ship image recognition through the model.

[0015] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent from the description, or may be understood by practicing the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which: Figure 1 is a flow chart of an infrared ship identification method according to an embodiment of the present application; Figure 2 It is a structural schematic diagram of an embodiment of the infrared ship identification system of the present application; Figure 3 It is a schematic diagram of the hardware structure of an embodiment of the electronic device of the present application. DETAILED DESCRIPTION

[0017] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as limiting the present application.

[0018] In the description of this application, if there is a description of first, second, etc., it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0019] In the description of the present application, it should be understood that the descriptions involving orientation, such as the orientation or positional relationship indicated as up, down, etc., are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0020] In the description of this application, it should be noted that, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meaning of the above terms in this application based on the specific content of the technical solution.

[0021] The technical solution of the present application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described below are only part of the embodiments of the present application, not all of the embodiments.

[0022] Figure 1 is a flow chart of an infrared ship identification method according to an embodiment of the present application; Figure 2 It is a structural schematic diagram of an embodiment of the infrared ship identification system of the present application; Figure 3 It is a schematic diagram of the hardware structure of an embodiment of the electronic device of the present application.

[0023] See below Figure 1 , the embodiments of the present application are further described.

[0024] The present application embodiment proposes an infrared ship recognition method, which includes the following steps: Step 101: Acquire an infrared ship image to be identified; Step 102: input the infrared ship image to be identified into the trained student model to obtain the recognition result of the infrared ship image; The trained student model is obtained by the following steps: Acquire a clean infrared ship sample, wherein the clean infrared ship sample includes a plurality of identifiable clean infrared ship images; Acquire a random noise vector group, wherein the random noise vector group includes a plurality of random noise vectors; According to the clean infrared ship samples and the random noise vector group, an enhanced infrared ship sample is obtained, wherein the enhanced infrared ship sample has the same characteristic distribution as the clean infrared ship sample; The student model is trained according to the clean infrared ship samples, the enhanced infrared ship samples and the teacher model that has completed adversarial training to obtain a trained student model.

[0025] In an embodiment of the present application, a student model is trained to obtain a trained student model suitable for infrared ship recognition. This model is used for recognition to effectively improve the accuracy of infrared ship image recognition. In the training process of the student model, clean infrared ship samples, enhanced infrared ship samples generated from clean infrared ship samples, and a teacher model that has completed adversarial training are used to enhance the diversity of infrared ship images, improve the robustness and generalization ability of the model, and thereby improve the accuracy of infrared ship image recognition through the model.

[0026] In addition, it can improve the robustness of the model against attack samples containing disturbance information. After completing the model construction and corresponding adversarial training, it can improve the anti-interference ability in infrared ship image recognition, improve the recognition accuracy in military security, and defend against malicious disturbances and other human attacks. At the same time, it can also adapt to the interference of complex natural environments such as different sea surfaces and temperatures.

[0027] The clean infrared ship samples may include a plurality of identifiable clean infrared ship images that are actually collected and then screened or processed.

[0028] The above-mentioned random noise vector group, including multiple random noise vectors, can be generated by a random noise generator.

[0029] For example, a random number seed is uniformly distributed or normally distributed, and a random noise generator is used to generate a set of single-channel random noise vectors according to the characteristics of the infrared image. , n is the number of random noise vectors.

[0030] The random number seed mentioned above is a value used to initialize the state of the random number generator. By changing the random number seed, different data samples can be generated, which is very useful for evaluating the performance of the generator and exploring its capabilities. The uniform distribution mentioned above means that each possible value has the same probability of being selected. This feature enables the generator to explore extensively in the potential space, giving it the opportunity to generate more diverse outputs. The normal distribution mentioned above, many natural phenomena and processes follow a normal distribution, such as human height and weight. The normal distribution as a source of random noise can better match the statistical characteristics of the target data. The infrared image characteristics mentioned above include: temperature gradient, the radiation intensity of different temperature areas in the infrared image is significantly different, and the generator needs to be able to accurately simulate these temperature gradients; dynamic range under low light, infrared images usually have a higher dynamic range, and the generator needs to be able to generate images containing high brightness and low brightness areas; colorless characteristics, infrared images are grayscale images, and the generator needs to generate high-quality grayscale images. In general, single-channel noise can be used here to simulate the various characteristics of infrared images.

[0031] The above-mentioned random noise generator, in the generative adversarial network, usually starts from a random noise vector. This noise vector can come from a uniform distribution or a normal distribution. What the generator has to do is to use this single-channel random noise vector to fit the clean sample to achieve the training of the generator.

[0032] In some implementations, the infrared ship images to be identified may also be used to continuously optimize the trained student model and update the model regularly.

[0033] In some implementations, training a student model based on clean infrared ship samples, enhanced infrared ship samples, and a teacher model that has completed adversarial training to obtain a trained student model includes: Inputting the training samples into the teacher model that has completed adversarial training to obtain a first output distribution result, wherein the training samples include clean infrared ship samples and enhanced infrared ship samples, and the teacher model that has completed adversarial training is obtained by inputting the adversarial samples into the teacher model to complete adversarial training; Input the training samples into the student model to obtain the second output distribution result; Input the training samples and adversarial samples into the student model to obtain the third output distribution result; The student model is trained according to the first output distribution result, the second output distribution result and the third output distribution result to obtain a trained student model.

[0034] The first output distribution result can be the output distribution obtained by inputting the teacher model with a mixture of clean infrared ship samples, enhanced infrared ship samples, and adversarial samples generated during adversarial training. The above second output distribution result can be the output distribution obtained after the clean infrared ship samples and the enhanced infrared ship samples are input into the student model. The third output distribution result can be the output distribution obtained by mixing clean infrared ship samples, enhanced infrared ship samples and adversarial samples into the student model. .

[0035] The above-mentioned training of the student model according to the first output distribution result, the second output distribution result and the third output distribution result to obtain the trained student model can be the process of knowledge distillation, and the first output distribution result of the teacher model that has completed the adversarial training is obtained. At temperature Under the condition of smooth soft labels, the second output distribution results of the student model are supervised And the third output distribution results , obtain the corresponding distillation labels and train the student model.

[0036] The above-mentioned adversarial training is a crucial technology for improving the robustness of machine learning models. The core idea of ​​adversarial training is to enhance the robustness of the model by introducing adversarial examples. Adversarial examples refer to small but carefully designed perturbations to the original input data, which cause the model's prediction results to be wrong. These perturbations are usually imperceptible to humans, but may cause serious misclassification in machine learning models. By adding these adversarial examples during the training process, the model can learn more robust feature representations, so that it can perform better in the face of similar attacks. It plays an important role in image recognition, natural language processing and security protection. In image recognition tasks, adversarial training helps models better resist adversarial attacks and improve safety in practical applications. For example, autonomous driving systems need to be able to identify and respond to visual interference that may be created by malicious attackers. In natural language processing, adversarial training improves the robustness of the model to text data and prevents the model from making incorrect predictions due to small text changes. This is particularly important for tasks such as sentiment analysis and machine translation. In the field of network security, adversarial training is used to detect and defend against malware and network attacks. By simulating the attacker's strategy, the model can learn how to identify and block potential threats. In addition, depending on whether the structural information of the model can be obtained, adversarial attacks are divided into white-box attacks that can obtain model structural information and black-box attacks that target raw data input and target model output.

[0037] The core idea of ​​the Generative Adversarial Network is to generate realistic data samples through a game between two neural networks, the generator and the discriminator. The two networks compete with each other during the training process and eventually reach a state of equilibrium, allowing the generator to generate samples that are difficult to distinguish from real data. It has shown great application potential in many fields, including but not limited to image generation: generating high-resolution images, artworks, faces, etc.; video generation: generating realistic video clips for filmmaking, virtual reality, etc.; text generation: generating natural language text for automatic writing, dialogue systems, etc.; data enhancement: improving the performance of machine learning models by generating additional training data; medical imaging: generating high-quality medical images to assist diagnosis and treatment.

[0038] The above teacher model and student model are related to knowledge distillation technology, which is a technology that transfers the knowledge of a complex model, namely the teacher model, to a simple model, namely the student model. Through knowledge distillation, the scale and computational cost of the model can be significantly reduced while maintaining the performance of the model, making it more suitable for deployment in resource-constrained environments. Specifically, the soft labels generated by the teacher model during training are used as supervision signals for the student model. The soft labels contain not only the predicted probability of each category, but also the relative relationship between categories, which provides more information than traditional hard labels. In this way, the student model can learn the implicit knowledge of the teacher model, not just the final classification result.

[0039] In this implementation, a teacher model that has completed adversarial training is used, and the generator added during the adversarial training process effectively measures the difference in data distribution. The generated samples have the advantages of sharp edges and greater diversity, thereby improving the recognition accuracy of the model. The prediction of the teacher model after adversarial training is used as a supervision signal to smooth the problem of loss of true sample distribution caused by binary hard labels, thereby improving the robustness of the model. Therefore, by applying the adversarial generation capability and knowledge distillation teacher network guidance, efficient and stable adversarial training of the model can be completed.

[0040] In some implementations, obtaining enhanced infrared ship samples according to clean infrared ship samples and a random noise vector group includes: Inputting the i-th random noise vector and the clean infrared ship sample in the random noise vector group into the generator model, obtaining an intermediate infrared ship sample corresponding to the i-th random noise vector, wherein the intermediate infrared ship sample includes multiple intermediate infrared ship images, and i is a positive integer; Determine a first loss situation corresponding to the ith random noise vector according to the intermediate infrared ship samples and the clean infrared ship samples corresponding to the ith random noise vector; When the first loss situation does not meet the first training stop condition of the generator model, adjust the parameters of the generator model to obtain an updated generator model, update i to i+1, and return to input the i-th random noise vector in the random noise vector group and the clean infrared ship sample into the generator model until the first training stop condition is met to obtain a generator model that has completed training; Obtain an optimal noise vector, where the optimal noise vector is a random noise vector in the random noise vector group; The optimal noise vector and clean infrared ship samples are input into the trained generator model to obtain enhanced infrared ship samples.

[0041] In this embodiment, clean infrared ship samples and random noise vector groups are passed through a trained generator model to generate enhanced infrared ship samples, which can enhance the quantity and diversity of original clean infrared ship samples, solve the problem of scarce high-quality infrared ship images used to train models, and the enhanced infrared ship samples are obtained through the optimal noise vector, which can further improve the quality of the enhanced infrared ship samples.

[0042] The generator model can be a deep convolutional generative adversarial network, which uses a convolutional neural network as the basic architecture of the generator, is suitable for generating high-quality images, and includes a convolutional layer, batch normalization, a fully connected layer, etc. Specifically, the generator model can receive a random noise vector as input and output a single-channel image with a shape of (64, 64, 1), that is, an infrared image.

[0043] The first training stop condition may be for each random noise vector until all random noise vectors can generate images that meet the requirements through the generator, so that the generator model can use any random noise to generate enhanced infrared ship samples that meet the distribution of clean infrared ship samples. Specifically, the first loss condition may be less than a preset value or the model converges, and the first loss condition may be an expected loss value.

[0044] For example, the loss value is expected to be constrained by the following mathematical expression: ; in, It represents the calculation of the expectation of the sample x drawn from the clean sample distribution X. is the generator after t steps of generative adversarial network training. is a single-channel random noise vector Random noise in ; It can be expressed as finding the optimal noise among all possible noise vectors Make the generator The generated image The Euclidean distance between the clean sample x is the smallest, that is, the difference between the two vectors is the smallest.

[0045] In some embodiments, the optimal noise vector is constrained by the following mathematical expression: ; in, is the optimal noise vector, Z is the random noise vector group, is the generator model that has completed training, and x is a clean infrared ship sample.

[0046] In this implementation, the optimal noise vector can be determined. The number of noise vectors is used as the loop variable for loop training. For each group of random noise, until all noise vectors can generate images that meet the requirements through the generator, the corresponding noise that best matches the original clean infrared ship sample is retained. argmin represents the parameter that minimizes the objective function. The above formula indicates that among all possible random noise vectors z, an optimal z is found, that is, , so that the image generated by the generator model With real image The Euclidean distance between them is the smallest, that is, the similarity is the largest.

[0047] In some implementations, a discriminator model can be added to accelerate the convergence of the generator model by feeding back the classification results of the discriminator model. After passing through the generator as input Train the discriminant classifier to obtain the corresponding enhanced infrared ship samples , is the output image sample, is the label corresponding to the image sample.

[0048] In some embodiments, training a student model according to the first output distribution result, the second output distribution result, and the third output distribution result includes: Obtaining a second loss situation according to the first output distribution result, the second output distribution result, and the third output distribution result; When the second loss situation does not meet the second training stop condition of the student model, the parameters of the student model are adjusted to obtain an updated student model, and the second loss situation is obtained according to the first output distribution result, the second output distribution result and the third output distribution result until the second training stop condition is met to obtain a student model that has completed training.

[0049] In this implementation, through knowledge distillation, the student model can approach or even exceed the performance of the teacher model as much as possible while maintaining a small model size and low computing resource requirements. By learning the complex decision-making process and knowledge of the teacher model, the student model can demonstrate better generalization and robustness on certain tasks. Through the guidance of the teacher model, the student model can effectively learn and inherit the knowledge and experience of the teacher model.

[0050] The second training stop condition may be that the second loss is less than a preset value, or the model converges.

[0051] In some embodiments, the second loss condition is constrained by the following mathematical expression: ; Among them, S is the student model, T is the teacher model that has completed adversarial training, is an adjustable parameter between 0 and 1. Represents the first output distribution result and the second output distribution result Divergence, Represents the second output distribution result and the third output distribution result Divergence.

[0052] In this embodiment, the first output distribution result and the second output distribution result are comprehensively considered. The divergence and the second output distribution result and the third output distribution result Divergence, clean infrared ship samples, enhanced infrared ship samples and adversarial samples are used to train the model, which can effectively improve the robustness of the model.

[0053] Above Divergence, for and Two data distributions, and Between The divergence can be constrained by the following formula: ; Specifically, for , is the second output distribution result, is the first output distribution result, for , is the second output distribution result, is the third output distribution result.

[0054] In some embodiments, adversarial examples are constrained by the following mathematical expression: ; in, is an adversarial sample, x is a clean infrared ship sample, is the preset disturbance value, is a constraint condition, representing an adversarial sample With the original sample The difference between The disturbance value does not exceed the preset value. , f is a parameter The adversarial generation sample model, It represents the maximum output error of the adversarial sample after passing through the model, and y is the true label of the clean infrared ship sample.

[0055] In this implementation, adversarial samples can be generated by the above formula. Adversarial samples can be reflected in the interference of a large amount of disturbance information in a complex background on the recognition accuracy of the model during infrared ship detection. Adversarial training is to consider such low-quality samples and train them together with ideal clean samples without interference, so as to enable the model to cope with different environments.

[0056] The above norm p can adopt L1 norm, L2 norm and L∞ norm. L1 norm promotes sparse solutions, is robust to outliers but is non-smooth and may fall into local optimum; L2 norm produces non-sparse solutions, is numerically stable and easy to optimize, but is sensitive to outliers; L∞ norm focuses on the largest element and is suitable for evaluating the worst case, but performs poorly in other information utilization and sparsity. The type of norm can be selected according to the needs of the specific application scenario, such as feature selection, model stability, and sensitivity to outliers.

[0057] The infrared ship recognition method provided in the embodiment of the present application may be executed by the infrared ship recognition system 200. In the embodiment of the present application, the infrared ship recognition system 200 executing the infrared ship recognition method is taken as an example to illustrate the infrared ship recognition system 200 provided in the embodiment of the present application.

[0058] See also Figure 2 , is a schematic diagram of the structure of an infrared ship identification system 200 provided in an embodiment of the present application. Figure 2 As shown, the infrared ship identification system 200 includes: An acquisition module 201 is used to acquire an infrared ship image to be identified; The recognition module 202 is used to input the infrared ship image to be recognized into the trained student model to obtain the recognition result of the infrared ship image; The trained student model is obtained by the following steps: Acquire a clean infrared ship sample, wherein the clean infrared ship sample includes a plurality of identifiable clean infrared ship images; Acquire a random noise vector group, wherein the random noise vector group includes a plurality of random noise vectors; According to the clean infrared ship samples and the random noise vector group, an enhanced infrared ship sample is obtained, wherein the enhanced infrared ship sample has the same characteristic distribution as the clean infrared ship sample; The student model is trained according to the clean infrared ship samples, the enhanced infrared ship samples and the teacher model that has completed adversarial training to obtain a trained student model.

[0059] In some implementations, the identification module 202 may be used to: Inputting the training samples into the teacher model that has completed adversarial training to obtain a first output distribution result, wherein the training samples include clean infrared ship samples and enhanced infrared ship samples, and the teacher model that has completed adversarial training is obtained by inputting the adversarial samples into the teacher model to complete adversarial training; Input the training samples into the student model to obtain the second output distribution result; Input the training samples and adversarial samples into the student model to obtain the third output distribution result; The student model is trained according to the first output distribution result, the second output distribution result and the third output distribution result to obtain a trained student model.

[0060] In some implementations, the identification module 202 may be used to: Inputting the i-th random noise vector and the clean infrared ship sample in the random noise vector group into the generator model, obtaining an intermediate infrared ship sample corresponding to the i-th random noise vector, wherein the intermediate infrared ship sample includes multiple intermediate infrared ship images, and i is a positive integer; Determine a first loss situation corresponding to the ith random noise vector according to the intermediate infrared ship samples and the clean infrared ship samples corresponding to the ith random noise vector; When the first loss situation does not meet the first training stop condition of the generator model, adjust the parameters of the generator model to obtain an updated generator model, update i to i+1, and return to input the i-th random noise vector in the random noise vector group and the clean infrared ship sample into the generator model until the first training stop condition is met to obtain a generator model that has completed training; Obtain an optimal noise vector, where the optimal noise vector is a random noise vector in the random noise vector group; The optimal noise vector and clean infrared ship samples are input into the trained generator model to obtain enhanced infrared ship samples.

[0061] In some embodiments, the optimal noise vector is constrained by the following mathematical expression: ; in, is the optimal noise vector, Z is the random noise vector group, is the generator model that has completed training, and x is a clean infrared ship sample.

[0062] In some implementations, the identification module 202 may be used to: Obtaining a second loss situation according to the first output distribution result, the second output distribution result, and the third output distribution result; When the second loss situation does not meet the second training stop condition of the student model, the parameters of the student model are adjusted to obtain an updated student model, and the second loss situation is obtained according to the first output distribution result, the second output distribution result and the third output distribution result until the second training stop condition is met to obtain a student model that has completed training.

[0063] In some embodiments, the second loss condition is constrained by the following mathematical expression: ; Among them, S is the student model, T is the teacher model that has completed adversarial training, is an adjustable parameter between 0 and 1. Represents the first output distribution result and the second output distribution result Divergence, Represents the second output distribution result and the third output distribution result Divergence.

[0064] In some embodiments, adversarial examples are constrained by the following mathematical expression: ; in, is an adversarial sample, x is a clean infrared ship sample, is the preset disturbance value, is a constraint condition, representing an adversarial sample With the original sample The difference between The disturbance value does not exceed the preset value. , f is a parameter The adversarial generation sample model, It represents the maximum output error of the adversarial sample after passing through the model, and y is the true label of the clean infrared ship sample.

[0065] Since the infrared ship recognition system 200 adopts all the technical solutions of the infrared ship recognition method of the above embodiment, it has at least all the beneficial effects brought by the technical solutions of the above embodiment, which will not be described in detail here.

[0066] Figure 3 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application.

[0067] The electronic device may include a processor 301 and a memory 302 storing computer program instructions.

[0068] Specifically, the processor 301 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0069] The memory 302 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 302 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. Where appropriate, the memory 302 may include a removable or non-removable (or fixed) medium. Where appropriate, the memory 302 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 302 is a non-volatile solid-state memory.

[0070] In some embodiments, the memory 302 may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage media device, an optical storage media device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Thus, in general, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to an aspect of the present disclosure.

[0071] The processor 301 implements any one of the infrared ship recognition methods in the above embodiments by reading and executing the computer program instructions stored in the memory 302.

[0072] In one example, the electronic device may further include a communication interface 303 and a bus 310. Figure 3 As shown, the processor 301, the memory 302, and the communication interface 303 are connected via a bus 310 and communicate with each other.

[0073] The communication interface 303 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0074] Bus 310 includes hardware, software or both, and couples the components of online data traffic billing equipment to each other. For example, but not limitation, the bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front-side bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnect (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. Where appropriate, bus 310 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the present application considers any suitable bus or interconnection.

[0075] The electronic device can execute the infrared ship recognition method in the embodiment of the present application, thereby realizing the combination Figure 1 and Figure 2 An infrared ship recognition method and system are described.

[0076] In addition, in combination with the infrared ship recognition method in the above embodiment, the embodiment of the present application can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any infrared ship recognition method in the above embodiment is implemented.

[0077] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0078] The functional blocks shown in the structural block diagram described above can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present application are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier. "Machine-readable medium" may include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROMs, flash memories, erasable ROMs (EROMs), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0079] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.

[0080] Aspects of the present disclosure are described above with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the functions / actions specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0081] The above is only a specific implementation of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.

Claims

1. An infrared ship identification method, characterized in that: include: Acquire infrared ship images to be identified; Inputting the infrared ship image to be identified into the trained student model to obtain a recognition result of the infrared ship image; The student model that has completed training is obtained by the following steps: Acquire a clean infrared ship sample, wherein the clean infrared ship sample includes a plurality of identifiable clean infrared ship images; Acquire a random noise vector group, wherein the random noise vector group includes a plurality of random noise vectors; Acquire enhanced infrared ship samples according to the clean infrared ship samples and the random noise vector group, wherein the enhanced infrared ship samples have the same characteristic distribution as the clean infrared ship samples; The student model is trained according to the clean infrared ship samples, the enhanced infrared ship samples and the teacher model that has completed adversarial training to obtain the trained student model.

2. The method according to claim 1, characterized in that The step of training a student model according to the clean infrared ship samples, the enhanced infrared ship samples and the teacher model that has completed adversarial training to obtain the trained student model includes: Inputting the training samples into the teacher model that has completed adversarial training to obtain a first output distribution result, wherein the training samples include the clean infrared ship samples and the enhanced infrared ship samples, and the teacher model that has completed adversarial training is obtained by inputting adversarial samples into the teacher model to complete adversarial training; Input the training samples into the student model to obtain the second output distribution result; Inputting the training sample and the adversarial sample into the student model to obtain a third output distribution result; The student model is trained according to the first output distribution result, the second output distribution result and the third output distribution result to obtain the trained student model.

3. The method according to claim 1, characterized in that The step of obtaining enhanced infrared ship samples according to the clean infrared ship samples and the random noise vector group includes: Inputting the i-th random noise vector in the random noise vector group and the clean infrared ship sample into the generator model to obtain an intermediate infrared ship sample corresponding to the i-th random noise vector, wherein the intermediate infrared ship sample includes a plurality of intermediate infrared ship images, and i is a positive integer; Determine a first loss situation corresponding to the i-th random noise vector according to the intermediate infrared ship samples and the clean infrared ship samples corresponding to the i-th random noise vector; In the case where the first loss situation does not satisfy the first training stop condition of the generator model, adjusting the parameters of the generator model to obtain the updated generator model, and updating i to i+1, returning to the step of inputting the i-th random noise vector in the random noise vector group and the clean infrared ship sample into the generator model until the first training stop condition is satisfied, thereby obtaining the generator model that has completed training; Acquire an optimal noise vector, where the optimal noise vector is one of the random noise vectors in the random noise vector group; The optimal noise vector and the clean infrared ship sample are input into the generator model that has completed training to obtain the enhanced infrared ship sample.

4. The method according to claim 3, characterized in that The optimal noise vector is constrained by the following mathematical expression: ； in, is the optimal noise vector, Z is the random noise vector group, is the generator model that has completed training, and x is the clean infrared ship sample.

5. The method according to claim 2, characterized in that: The step of training the student model according to the first output distribution result, the second output distribution result and the third output distribution result comprises: Obtaining a second loss situation according to the first output distribution result, the second output distribution result, and the third output distribution result; When the second loss situation does not meet the second training stop condition of the student model, adjust the parameters of the student model to obtain the updated student model, return to the first output distribution result, the second output distribution result and the third output distribution result to obtain the second loss situation until the second training stop condition is met to obtain the student model that has completed training.

6. The method according to claim 5, characterized in that The second loss condition is governed by the following mathematical expression: ； Where S is the student model, T is the teacher model that has completed adversarial training, is an adjustable parameter between 0 and 1. Indicates the first output distribution result and the second output distribution result Divergence, Indicates the second output distribution result and the third output distribution result Divergence.

7. The method according to claim 2, characterized in that The adversarial example is constrained by the following mathematical expression: ； in, is the adversarial sample, x is the clean infrared ship sample, is the preset disturbance value, is a constraint condition, representing an adversarial sample With the original sample The difference between The preset disturbance value is not exceeded. , f is a parameter The adversarial generation sample model, It represents maximizing the output error of the adversarial sample after passing through the model, and y is the true label of the clean infrared ship sample.

8. An infrared ship identification system, characterized in that: include: An acquisition module, used for acquiring infrared ship images to be identified; A recognition module, used for inputting the infrared ship image to be recognized into the trained student model to obtain a recognition result of the infrared ship image; The student model that has completed training is obtained by the following steps: Acquire a clean infrared ship sample, wherein the clean infrared ship sample includes a plurality of identifiable clean infrared ship images; Acquire a random noise vector group, wherein the random noise vector group includes a plurality of random noise vectors; Acquire enhanced infrared ship samples according to the clean infrared ship samples and the random noise vector group, wherein the enhanced infrared ship samples have the same characteristic distribution as the clean infrared ship samples; The student model is trained according to the clean infrared ship samples, the enhanced infrared ship samples and the teacher model that has completed adversarial training to obtain the trained student model.

9. An electronic device, characterized in that: It comprises a processor and a memory, wherein the memory stores programs or instructions that can be run on the processor, and when the programs or instructions are executed by the processor, the steps of the infrared ship identification method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the infrared ship recognition method as described in any one of claims 1 to 7.

Images