Ship target image augmentation method, device and system based on conditional diffusion model

Abstract

The present application relates to a kind of ship target image expansion method, device and system based on conditional diffusion model.The method part mainly includes: obtaining ship target image data set, and determining the corresponding feature description text corresponding to ship target image;Based on ship target image and feature description text, train image text description model, and train conditional generation diffusion model;Using the feature description text of the ship target image to be generated as model condition, corresponding feature description text is vectorized based on image text description model;Again using the vectorized feature description text as model condition, using conditional generation diffusion model is predicted, obtains the ship target image generated after.This application proposes the image amplification method with the corresponding feature description text of ship target image as the restriction condition, by constructing image text description model, conditional generation diffusion model, to realize the automatic generation amplification of a large number of available ship target images meeting the demand.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image data set expansion, in particular to a ship target image expansion method, device and system based on a conditional diffusion model. BACKGROUND

[0002] In recent years, deep learning-based methods have achieved good performance in automatic driving, smart cities and many other fields, which is largely dependent on massive data. In some conditions, data in some fields is difficult to obtain due to technical or cost reasons, such as semantic segmentation data used in automatic driving, multi-angle multi-style face data, etc. The problems of insufficient image data samples, difficulty in obtaining, and inaccurate labeling have become one of the main problems hindering the development of artificial intelligence in practical application scenarios. In ship applications, in order to better analyze the target characteristics of both ships, a large number of ship image data covering various scenarios are usually required, which are usually difficult to collect. Image augmentation is an effective way to solve this problem. By automatically expanding data through intelligent methods, a diffusion model is a widely used image generation model that can generate similar pictures in the field based on a large amount of unsupervised image data in the field. Since the effect of generating ship pictures by this method is uncontrollable, it is difficult to directly generate usable ship pictures that meet the requirements.

[0003] Therefore, how to overcome the defects of the prior art and solve the problem of uncontrollable effect of generating ship pictures has become an important technical problem to be solved in the industry. SUMMARY

[0004] In view of the defects or improvement needs in the prior art: the effect of generating ship pictures is uncontrollable, and it is difficult to directly generate usable ship pictures that meet the requirements. The present application provides a ship target image expansion method, device and system based on a conditional diffusion model, which proposes an image expansion method with the corresponding feature description text of the ship target image as the limiting condition, and realizes the automatic generation and expansion of a large number of usable ship target images that meet the requirements by constructing an image text description model and a conditional generation diffusion model.

[0005] The present application adopts the following technical solutions:

[0006] In a first aspect, the present application provides a ship target image expansion method based on a conditional diffusion model, comprising:

[0007] obtaining a ship target image data set and determining the corresponding feature description text corresponding to the ship target image;

[0008] training an image text description model based on the ship target image and the feature description text, and training a conditional generation diffusion model;

[0009] The feature description text of the ship target image to be generated is used as a model condition, and the corresponding feature description text is vectorized based on an image text description model; the vectorized feature description text is used as a model condition, and a conditional generation diffusion model is used for prediction to obtain the generated ship target image.

[0010] In some embodiments, the training image text description model specifically comprises:

[0011] The input of the image text description model includes two parts of the ship target image and the feature description text, wherein the ship target image is encoded using an image encoder, and the feature description text is encoded using a text encoder; after the encoding of the two parts is completed, the similarity of the vector after the encoding of the image encoder and the vector after the encoding of the text encoder is calculated; the training target of the image text description model is to make the similarity of the positive example samples higher and the similarity of the negative example samples lower.

[0012] In some embodiments, the training conditional generation diffusion model specifically comprises:

[0013] For data x0 in the ship target image data set D, Gaussian noise is added to it through T times of accumulation to obtain x1, x2, …, x T , and the way of adding Gaussian noise at the t-th step (t e {1, 2, …, T}) is as follows:

[0014]

[0015] wherein ε t ~ N(0, I), β t is a noise intensity coefficient, satisfying 0 < β1< β2< … < β T <1 and α t + β t = 1, the noise intensity coefficient β t is set based on the following formula:

[0016]

[0017] Through the above noise diffusion process, the Gaussian noise matrix x T corresponding to the data x0 can be obtained.

[0018] In some embodiments, the use of the feature description text of the ship target image to be generated as a model condition and the vectorization of the corresponding feature description text based on the image text description model specifically comprises:

[0019] The feature description text of the ship target image to be generated is used as a model condition, and the corresponding feature description text is vectorized based on an image text description model; the vectorized feature description text is used as a model condition, and a conditional generation diffusion model is used for prediction to obtain the generated ship target image. The feature description text of the ship target image to be generated is used as a model condition, and the corresponding feature description text is vectorized based on an image text description model; the vectorized feature description text is used as a model condition, and a conditional generation diffusion model is used for prediction to obtain the generated ship target image.

[0020] In some embodiments, using vectorized feature description text as model conditions and using a conditional generation diffusion model for prediction to obtain the generated ship target image specifically includes:

[0021] Build a neural network ε θ Given (x, t, c), the network parameter is θ. The mean squared error is used to measure the network loss, and the network loss function is expressed as:

[0022]

[0023] in, c is the vectorized feature description text of the ship target image to be generated. The network is trained using gradient descent until the network converges.

[0024] x is obtained by sampling from the standard normal distribution. T , that is, x T ~N(0,I), based on the following formula, T iterations are performed, changing the mean of the Gaussian samples during the sampling process:

[0025]

[0026] in, Samples are generated after T iterations.

[0027] In some embodiments, the feature description text includes one or more of the following: different types of description text for ship target images, different model description texts, different operating status description texts, and different environmental condition description texts.

[0028] In some embodiments, the method further includes: reviewing the generated ship target image and incorporating the approved ship target image and the input feature description text into the original dataset.

[0029] Secondly, the present invention also provides a ship target image augmentation device based on a conditional diffusion model, the device comprising:

[0030] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor for performing the ship target image augmentation method based on the conditional diffusion model described in the first aspect.

[0031] Thirdly, the present invention also provides a ship target image augmentation system based on a conditional diffusion model, applying the ship target image augmentation method based on a conditional diffusion model as described in the first aspect. The system includes a dataset establishment module, a model training module, and a target image generation module, wherein:

[0032] The dataset creation module is used to acquire a dataset of ship target images and determine the corresponding feature description text for each ship target image.

[0033] The model training module trains an image-text description model and a conditional generation diffusion model based on ship target images and feature description text.

[0034] The target image generation module uses the feature description text of the ship target image to be generated as the model condition, and vectorizes the corresponding feature description text based on the image text description model; then, it uses the vectorized feature description text as the model condition and uses the conditional generation diffusion model to predict and obtain the generated ship target image.

[0035] Fourthly, the present invention also provides a non-volatile computer storage medium storing computer-executable instructions, which are executed by one or more processors to perform the ship target image augmentation method based on the conditional diffusion model described in the first aspect.

[0036] Compared with the prior art, the present invention provides a method, apparatus and system for augmenting ship target images based on a conditional diffusion model. Its advantages are as follows: it proposes an image augmentation method with the corresponding feature description text of the ship target image as a constraint condition. By constructing an image text description model and a conditional generation diffusion model, it can realize the automatic generation and augmentation of a large number of usable ship target images that meet the requirements. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments of the present invention will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0038] Figure 1 This is a flowchart of a ship target image augmentation method based on a conditional diffusion model provided in Embodiment 1 of the present invention;

[0039] Figure 2 A flowchart of the training image text description model provided in Embodiment 1 of the present invention;

[0040] Figure 3This is a flowchart of the process for generating a ship target image provided in Embodiment 1 of the present invention;

[0041] Figure 4 This is a schematic diagram of a ship target image augmentation system based on a conditional diffusion model provided in Embodiment 2 of the present invention;

[0042] Figure 5 This is a schematic diagram of a ship target image augmentation device based on a conditional diffusion model, provided in Embodiment 3 of the present invention. Detailed Implementation

[0043] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention. It should be noted that, unless otherwise specified, the various features in the embodiments of the present invention can be combined with each other, all within the protection scope of this application. Furthermore, although functional modules may be divided in the device schematic diagram, and a logical order may be shown in the flowchart, in some cases, the steps shown or described may be executed differently from the module division in the device or the order in the flowchart.

[0044] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of the invention described below can be combined with each other as long as they do not conflict with each other.

[0045] Unless the context otherwise requires, throughout the specification and claims, the term "comprising" is interpreted as openly inclusive, meaning "including, but not limited to." In the description of the specification, terms such as "one embodiment," "some embodiments," "exemplary embodiment," "example," "specific example," or "some examples" are intended to indicate that a particular feature, structure, material, or characteristic associated with that embodiment or example is included in at least one embodiment or example of this disclosure. The illustrative representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics mentioned may be included in any suitable manner in any one or more embodiments or examples; that is, although they may be incorporated into embodiments or examples using the above terms for reasons such as order and position, it does not limit them to be incorporated in combination by a single embodiment or example.

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. The invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0047] Example 1:

[0048] This invention aims to expand existing ship target images. The scheme first trains an image text description model, then constructs a conditional generation diffusion model that accepts input image generation constraints, and finally generates usable ship images, thus expanding the dataset. The overall method is as follows.

[0049] like Figure 1 As shown in the figure, this embodiment of the invention provides a method for augmenting ship target images based on a conditional diffusion model, which includes the following steps.

[0050] Step 101: Obtain the ship target image dataset and determine the corresponding feature description text for each ship target image. In this step, the dataset is acquired and labeled. A large number of ship target images are obtained, covering various ship target images of different types, models, and operating states. The ship target images are then labeled, which means labeling the corresponding feature description text.

[0051] Step 102: Train an image-text description model and a conditional generation-diffusion model based on ship target images and feature description text. In this step, the image-text description model is trained using labeled ship target image-feature description text pairs to generate intelligent models capable of producing ship image description text. The input to this model includes both the ship target image and the feature description text, which are encoded using an image encoder and a text encoder, respectively. The similarity between the encoded vectors of the two parts is then calculated, and the loss is calculated based on the similarity value for model training. Additionally, the conditional generation-diffusion model is also trained using a large number of labeled ship target images and feature description texts.

[0052] Step 103: Using the feature description text of the target ship image to be generated as the model condition, the corresponding feature description text is vectorized based on the image text description model; then, using the vectorized feature description text as the model condition, a conditional generation-diffusion model is used for prediction to obtain the generated target ship image. In this step, during prediction, the feature description text of the target ship image to be generated is used as the model condition, and the feature description text is vectorized based on the image text description model; finally, a conditional generation-diffusion model is used for prediction to obtain the generated target ship image, which is then reviewed and filtered before being incorporated into the original image dataset.

[0053] Through the above method, this invention provides a ship target image augmentation method based on a conditional diffusion model. It proposes an image augmentation method with corresponding feature description text of the ship target image as a constraint. By constructing an image text description model and a conditional generation diffusion model, a large number of usable ship target images that meet the requirements can be automatically generated and augmented.

[0054] The solution of this embodiment will be described in detail below.

[0055] In some embodiments, the feature description text includes one or more of the following: description text for different types of ship target images, description text for different models, description text for different operating states, and description text for different environmental conditions. Specifically, when acquiring and labeling the dataset, a large number of ship target images are acquired, including various ship images covering different types, models, operating states, and environmental conditions. These ship target images are then labeled with corresponding feature description texts. Before labeling, the attribute values ​​of various characteristics of the ship target images are first determined. For example, ship type may include "large ship," "small ship," etc.; ship operating state includes "moving fast," "berthed," "moving slowly," "turning," "launching equipment," etc.; environmental condition includes "good sea conditions," "poor sea conditions," etc. Each image should be labeled based on preset labeling requirements.

[0056] refer to Figure 2 As shown, in some embodiments, the training of the image text description model specifically includes the following steps.

[0057] Step 201: Input the ship target image and feature description text. Encode the ship target image using an image encoder and the feature description text using a text encoder.

[0058] Step 202: Calculate the similarity between the vector encoded by the image encoder and the vector encoded by the text encoder.

[0059] Step 203: Train the image-text description model so that the similarity of positive samples is higher and the similarity of negative samples is lower.

[0060] The above steps describe the training of an image-text description model. Based on labeled ship target image-feature description text pairs, an intelligent model capable of generating ship image description text is trained. The input to the image-text description model consists of two parts: the ship target image and the feature description text. The ship target image is encoded using an image encoder, and the feature description text is encoded using a text encoder. The image encoder can be a neural network structure such as ResNet, while the text encoder can use embeddings from language models (ELMO) or a bidirectional encoder representation from transformers (BERT) network. ELMO and BERT are both pre-training methods in natural language processing. After both parts are encoded, the similarity between the vector encoded by the image encoder and the vector encoded by the text encoder is calculated. The training objective of the image-text description model is to maximize the similarity of positive samples and minimize the similarity of negative samples.

[0061] In some embodiments, the step of using the feature description text of the ship target image to be generated as a model condition and vectorizing the corresponding feature description text based on the image text description model specifically includes: using the text encoder in the trained image text description model to encode the feature description text of the ship target image to be generated, and obtaining the encoded vector, which is the vectorized feature description text of the ship target image to be generated.

[0062] Next, we need to predict and generate ship target images. First, input the feature description text of the desired ship target image, and use the conditional generation diffusion model to predict and obtain the generated ship target image. The structure of the conditional generation diffusion model can be referenced from the Denoising Diffusion Probabilistic Models (DDPM). The model training and prediction process is described below.

[0063] In some embodiments, the training conditional generation diffusion model specifically includes: taking data x0 in the ship target image dataset D (each data x0 includes a ship target image and corresponding feature description text), adding Gaussian noise to it T times to obtain x1, x2, ..., x T The method for adding Gaussian noise in step t (t∈{1,2,…,T}) is as follows:

[0064]

[0065] Where, ε t ~N(0,I),β t Let β be the noise intensity coefficient, satisfying 0 < β1 < β2 < ... < β T <1 and α t +β t =1,β t There are many ways to take the value of β, for example, the noise intensity coefficient β t It can be set based on the following formula:

[0066]

[0067] Through the above noise diffusion process, the Gaussian noise matrix x corresponding to the data x0 can be obtained. T .

[0068] In some embodiments, using vectorized feature description text as model conditions and using a conditional generation diffusion model for prediction to obtain the generated ship target image specifically includes: building a neural network ε θ (x t Given a network with parameters θ (t, c), and using mean squared error to measure network loss, the network loss function can be expressed as:

[0069]

[0070] in, c represents the vectorized feature description text of the ship target image to be generated, including target type, target state, and environmental sea conditions. This feature description text constrains the training process of the conditional generation-diffusion model. The feature description text is encoded using the text encoder of the previously trained image text description model. The model is trained using gradient descent until convergence.

[0071] Finally, data is generated by sampling x from a standard normal distribution. T , that is, x T ~N(0,I), based on the following formula, T iterations are performed, changing the mean of the Gaussian samples during the sampling process:

[0072]

[0073] in, Samples are generated after T iterations. generated samples Similar to data x0 in dataset D above, each dataset includes a ship target image and corresponding feature description text.

[0074] Based on the specific process of generating ship target images described above, refer to Figure 3 As shown, the steps for generating a ship target image can be summarized as follows.

[0075] Step 301: Add Gaussian noise to the data in the ship target image dataset by accumulating it T times. Through the noise diffusion process, the corresponding Gaussian noise matrix can be obtained.

[0076] Step 302: Build a neural network. Based on the obtained Gaussian noise matrix, use mean squared error to measure the network loss, and train the network using gradient descent until the network converges.

[0077] Step 303: Generate data. Obtain the Gaussian noise matrix by sampling from the standard normal distribution. Based on changing the mean of the Gaussian samples during the sampling process, perform T iterations. After T iterations, generate the target sample.

[0078] In some embodiments, the method further includes: reviewing the generated ship target images, and incorporating the reviewed ship target images and the input feature description text into the original dataset. It should be noted that the review of the generated sample images can be performed using manual sampling methods, and the reviewed ship target images and the corresponding input feature description text are then incorporated into the original dataset D.

[0079] In summary, this invention provides a ship target image augmentation method based on a conditional diffusion model. It proposes an image augmentation method that uses the corresponding feature description text of the ship target image as a constraint. By constructing an image text description model and a conditional generation diffusion model, a large number of usable ship target images that meet the requirements can be automatically generated and augmented.

[0080] Example 2:

[0081] Based on the ship target image augmentation method based on the conditional diffusion model provided in Embodiment 1 above, Embodiment 2 of the present invention also provides a ship target image augmentation system based on the conditional diffusion model, referring to... Figure 4 As shown, the system includes a dataset establishment module, a model training module, and a target image generation module. Specifically: the dataset establishment module acquires a dataset of ship target images and determines the corresponding feature description text for each ship target image; the model training module trains an image-text description model and a conditional generation-diffusion model based on the ship target images and feature description text; the target image generation module uses the feature description text of the ship target image to be generated as model conditions, vectorizes the corresponding feature description text based on the image-text description model, and then uses the vectorized feature description text as model conditions to perform prediction using the conditional generation-diffusion model to obtain the generated ship target image.

[0082] The specific implementation process for each module can be found in the steps outlined above, and will not be repeated here.

[0083] In summary, this invention provides a ship target image augmentation system based on a conditional diffusion model. It proposes an image augmentation method that uses the corresponding feature description text of the ship target image as a constraint. By constructing an image text description model and a conditional generation diffusion model, a large number of usable ship target images that meet the requirements can be automatically generated and augmented.

[0084] Example 3:

[0085] Based on the ship target image augmentation method based on the conditional diffusion model provided in Embodiment 1 above, the present invention also provides a ship target image augmentation device based on the conditional diffusion model that can be used to implement the above method and system, such as... Figure 5 The diagram shown is a schematic representation of the device architecture according to an embodiment of the present invention. The ship target image augmentation device based on a conditional diffusion model in this embodiment includes one or more processors 21 and a memory 22. Figure 5 Take a processor 21 as an example.

[0086] Processor 21 and memory 22 can be connected via a bus or other means. Figure 5 Taking the example of a connection between China and Israel via a bus.

[0087] The memory 22, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the ship target image augmentation method based on the conditional diffusion model in Embodiment 1. The processor 21 executes various functional applications and data processing of the ship target image augmentation device based on the conditional diffusion model by running the non-volatile software programs, instructions, and modules stored in the memory 22, thereby realizing the ship target image augmentation method based on the conditional diffusion model in Embodiment 1.

[0088] Memory 22 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 22 may optionally include memory remotely located relative to processor 21, which can be connected to processor 21 via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0089] The program instructions / modules are stored in memory 22. When executed by one or more processors 21, they perform the ship target image augmentation method based on the conditional diffusion model described in Embodiment 1 above, for example, performing the above-described... Figures 1-3 The steps shown.

[0090] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0091] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0092] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software and a general-purpose hardware platform, or of course, using hardware. Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; under the concept of the present invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the present invention as described above. For the sake of brevity, they are not provided in detail; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for augmenting ship target images based on a conditional diffusion model, characterized in that, include: Obtain a dataset of ship target images and determine the corresponding feature description text for each ship target image; Based on ship target images and feature description text, an image-text description model is trained, and a conditional generation-diffusion model is also trained. The training of the conditional generation-diffusion model specifically includes: training the ship target image dataset... Data in By adding Gaussian noise to it T times, we obtain Step t ( The method for adding Gaussian noise is as follows: ,in, , The noise intensity coefficient satisfies and Noise intensity coefficient Based on the following formula: Data can be obtained through the above noise diffusion process. The corresponding Gaussian noise matrix ; The feature description text of the ship target image to be generated is used as the model condition. The corresponding feature description text is vectorized based on the image text description model. Then, the vectorized feature description text is used as the model condition, and the conditional generation diffusion model is used to predict and obtain the generated ship target image. The step of using the feature description text of the ship target image to be generated as the model condition and vectorizing the corresponding feature description text based on the image text description model specifically includes: using the text encoder in the trained image text description model to encode the feature description text of the ship target image to be generated, and obtaining the encoded vector, which is the vectorized feature description text of the ship target image to be generated; The process of using vectorized feature description text as model conditions and employing a conditional generation diffusion model for prediction to obtain the generated ship target image specifically includes: building a neural network. The network parameters are The mean squared error is used to measure the network loss, and the network loss function is expressed as: ,in, ; This is a vectorized feature description text of the ship target image to be generated. The network is trained using gradient descent until it converges; samples are obtained from a standard normal distribution. ,Right now Based on the following formula, T iterations are performed, changing the mean of the Gaussian samples during the sampling process: ,in, Samples are generated after T iterations. .

2. The ship target image augmentation method based on conditional diffusion model according to claim 1, characterized in that, The trained image text description model specifically includes: The input to the image-text description model consists of two parts: a ship target image and feature description text. The ship target image is encoded using an image encoder, and the feature description text is encoded using a text encoder. After both parts are encoded, the similarity between the vector encoded by the image encoder and the vector encoded by the text encoder is calculated. The training objective of the image-text description model is to maximize the similarity of positive samples and minimize the similarity of negative samples.

3. The ship target image augmentation method based on the conditional diffusion model according to claim 1 or 2, characterized in that, The feature description text includes one or more of the following: different types of description text for ship target images, different model description texts, different operating status description texts, and different environmental condition description texts.

4. The ship target image augmentation method based on the conditional diffusion model according to claim 1 or 2, characterized in that, Also includes: The generated ship target images are reviewed, and the approved ship target images are merged into the original dataset along with the input feature description text.

5. A device for augmenting ship target images based on a conditional diffusion model, characterized in that, The apparatus includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the processor for performing the ship target image augmentation method based on the conditional diffusion model according to any one of claims 1-4.

6. A ship target image augmentation system based on a conditional diffusion model, employing the ship target image augmentation method based on a conditional diffusion model as described in any one of claims 1-4, characterized in that, The system includes a dataset creation module, a model training module, and a target image generation module, wherein: The dataset creation module is used to acquire a dataset of ship target images and determine the corresponding feature description text for each ship target image. The model training module trains an image-text description model and a conditional generation diffusion model based on ship target images and feature description text. The target image generation module uses the feature description text of the ship target image to be generated as the model condition, and vectorizes the corresponding feature description text based on the image text description model; then, it uses the vectorized feature description text as the model condition and uses the conditional generation diffusion model to predict and obtain the generated ship target image.

7. A non-volatile computer storage medium, characterized in that, The computer storage medium stores computer-executable instructions, which are executed by one or more processors to perform the ship target image augmentation method based on the conditional diffusion model as described in any one of claims 1-4.

Images