Target detection methods, systems, devices, and media based on multi-source image fusion

Abstract

The application discloses a target detection method, system, device and medium based on multi-source image fusion, and relates to the technical field of target detection, to solve the problem of poor accuracy of a target detection result. The target detection method comprises the following steps: first, a first source image and a second source image obtained by shooting the same target are acquired; then, a modal conversion model is used to perform modal conversion on the first source image, to obtain a simulated second source image; finally, the second source image and the simulated second source image are taken as input, and a segmentation model is used to obtain a segmentation result, to complete target detection. The target detection method, system, device and medium based on multi-source image fusion provided by the application are used for target detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target detection, and in particular to a target detection method, system, device and medium based on multi-source image fusion. BACKGROUND

[0002] According to different input information, the existing target detection segmentation algorithm can be divided into single-source segmentation and multi-source segmentation. The single-source segmentation mainly designs a segmentation algorithm for distinguishing the target and the background according to the prior information such as the difference in gray scale, shape and background environment of the target, such as threshold method, fuzzy clustering, active contour and semantic segmentation based on convolutional neural network CNN. The multi-source segmentation mainly fuses the image information of different sources, so as to complementarily use the information of the target under different sensors, and then realizes target recognition and segmentation on the fused image by using the single-source segmentation method. However, the existing multi-source segmentation method fuses the image before image segmentation, which usually needs to perform image registration, and it is very difficult to perform spatial registration on the images of different sources, which will cause part of the information to be lost in the image fusion process, and then affect the utilization efficiency of multi-source information and the accuracy of the target detection result. SUMMARY

[0003] The purpose of the present application is to provide a target detection method, system, device and medium based on multi-source image fusion, which can improve the accuracy of the target detection result.

[0004] In order to achieve the above purpose, the present application provides the following technical scheme:

[0005] A target detection method based on multi-source image fusion, comprising:

[0006] obtaining a first source image and a second source image obtained by photographing the same target; the first source image and the second source image are obtained by photographing the target by using different sensors;

[0007] using a modal conversion model to perform modal conversion on the first source image to obtain a simulated second source image;

[0008] using the second source image and the simulated second source image as input, using a segmentation model to obtain a segmentation result, and completing target detection.

[0009] Compared with the prior art, in the target detection method provided by the application, a first source image and a second source image are first acquired, then a modal conversion model is used to perform modal conversion on the first source image to obtain a simulated second source image, finally, the second source image and the simulated second source image are taken as input, and a segmentation model is used to obtain a segmentation result, so that target detection is completed. The target detection method of the application converts the first source image into the simulated second source image, then fuses the simulated second source image with the second source image, and does not involve the means of fusing different source images by spatial registration, so that the problem that part of information is lost in the image fusion process and the utilization efficiency of multi-source information is affected is not caused, the utilization efficiency of multi-source information can be improved by performing modal conversion between different source images, and the multi-source information can be complementary, and the precision of the target detection result is improved.

[0010] The application further provides a target detection system based on multi-source image fusion, comprising:

[0011] An acquisition module is configured to acquire a first source image and a second source image obtained by photographing a same target by using different sensors;

[0012] A conversion module is configured to perform modal conversion on the first source image by using a modal conversion model to obtain a simulated second source image;

[0013] A segmentation module is configured to take the second source image and the simulated second source image as input, and obtain a segmentation result by using a segmentation model, so that target detection is completed.

[0014] Compared with the prior art, the target detection system provided by the application has the same beneficial effects as the target detection method described in the above technical solution, and details are not repeated here.

[0015] The application further provides a target detection device based on multi-source image fusion, comprising:

[0016] A processor; and

[0017] A memory in which computer readable program instructions are stored,

[0018] When the computer readable program instructions are executed by the processor, the above target detection method is executed.

[0019] The application further provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the steps of the above target detection method. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:

[0021] Figure 1 A method flowchart of the target detection method provided in Embodiment 1 of the present application;

[0022] Figure 2 A structural schematic diagram of the modal conversion network provided in Embodiment 1 of the present application;

[0023] Figure 3 A structural schematic diagram of the dual-channel segmentation network provided in Embodiment 1 of the present application;

[0024] Figure 4 A system block diagram of the target detection system provided in Embodiment 2 of the present application. DETAILED DESCRIPTION

[0025] In order to clearly describe the technical solutions of the embodiments of the present application, in the embodiments of the present application, the same items or similar items with basically the same functions and effects are distinguished by using the words “first”, “second”, etc. For example, the first threshold value and the second threshold value are only used to distinguish different threshold values, and do not limit the order. Those skilled in the art can understand that the words “first”, “second”, etc. do not limit the quantity and execution order, and the words “first”, “second”, etc. also do not necessarily mean different.

[0026] It should be noted that in the present application, the words “exemplary” or “for example” are used to mean by way of example, illustration, or description. Any embodiment or design solution described in the present application as “exemplary” or “for example” should not be interpreted as more preferred or more advantageous than other embodiments or design solutions. Rather, the words “exemplary” or “for example” are used in the sense of presenting related concepts in a specific manner.

[0027] Embodiment 1:

[0028] This embodiment is used to provide a target detection method based on multi-source image fusion, as shown in Figure 1 The target detection method comprises the following steps:

[0029] S1: obtaining a first source image and a second source image obtained by photographing the same target; the first source image and the second source image are obtained by photographing the target using different sensors;

[0030] Specifically, the target is photographed by the first source sensor at a first angle to obtain a first source image. The target is photographed by the second source sensor at a second angle to obtain a second source image. It should be noted that the embodiment only needs to ensure that the target photographed by the first source sensor and the second source sensor is the same target, and does not make any limitation on other shooting conditions. For example, the first angle and the second angle can be the same or different, and the first source sensor and the second source sensor are arranged arbitrarily.

[0031] The first source sensor and the second source sensor of the embodiment can be, but are not limited to, isospectral heterogeneous sensors, such as visible light sensors and thermal infrared sensors, to perform target detection under dual-source images; heterogeneous heterogeneous sensors, such as image sensors and radar signal sensors, to perform target detection in a scene through the combination of images and radar signals. More specifically, the first source image is a visible light image, and the second source image is a thermal infrared image; or, the first source image is a thermal infrared image, and the second source image is a visible light image.

[0032] S2: performing modal conversion on the first source image by using a modal conversion model to obtain a simulated second source image;

[0033] The modal conversion model of the embodiment is obtained by training a modal conversion network, and the modal conversion network as a whole is a cycle generative adversarial network CycleGAN. As shown in the figure, Figure 2 The modal conversion network includes a first generator G1, a second generator G2, a first discriminator D1, and a second discriminator D2. The first generator G1 is connected with the first discriminator D1 and the second generator G2 respectively, and the second generator G2 is further connected with the second discriminator D2.

[0034] Before S2, the target detection method of the embodiment further includes a step of training the modal conversion network to obtain the modal conversion model, which can include:

[0035] obtaining a first training sample set; the first training sample set includes a plurality of training first source images and a plurality of training second source images;

[0036] For each training first source image, the first generator G1 is used to perform modal conversion on the training first source image to obtain a training simulated second source image;

[0037] The second generator G2 is used to perform modal conversion on the training simulated second source image to obtain a training reconstructed first source image, and a first pixel-level loss between the training reconstructed first source image and the training first source image is calculated;

[0038] According to the training simulated second source image and any training second source image, the first discriminator D1 is used to obtain a first GAN loss;

[0039] For each training second source image, a training simulated first source image is obtained by modal conversion of the training second source image using the second generator G2;

[0040] A training reconstructed second source image is obtained by modal conversion of the training simulated first source image using the first generator G1, and a second pixel-level loss between the training reconstructed second source image and the training second source image is calculated;

[0041] A second GAN loss is obtained by the second discriminator D2 according to the training simulated first source image and any training first source image;

[0042] A first twin semantic loss between an intermediate feature of the first generator G1 when converting the training first source image to obtain the training simulated second source image and an intermediate feature of the first generator G1 when converting the training simulated first source image to obtain the training reconstructed second source image is calculated, and a second twin semantic loss between an intermediate feature of the second generator G2 when converting the training second source image to obtain the training simulated first source image and an intermediate feature of the second generator G2 when converting the training simulated second source image to obtain the training reconstructed first source image is calculated;

[0043] A loss value is calculated according to the first pixel-level loss, the first GAN loss, the second pixel-level loss, the second GAN loss, the first twin semantic loss and the second twin semantic loss, and whether a first iteration end condition is reached is determined according to the loss value;

[0044] If yes, the iteration is ended, and the first generator G1 used in the current iteration is taken as the modal conversion model; that is, the modal conversion model only includes the trained first generator G1 to convert the first source image into the simulated second source image;

[0045] If no, the iteration is continued, the parameters of the first generator G1, the second generator G2, the first discriminator D1 and the second discriminator D2 are adjusted according to the loss value, and the step of “for each training first source image, the training first source image is modal converted using the first generator G1” is returned.

[0046] The embodiment introduces the first twin semantic loss of the first generator G1 and the second twin semantic loss of the second generator G2 in the training process of the modal conversion network to strengthen the information correlation degree of the modal conversion process. After the first source image is converted into the simulated second source image by using the modal conversion model trained by the training process, the simulated second source image has the advantages and characteristics of the high resolution and high definition of the first source image, the modal conversion effect is good, the first source image can be converted into the simulated second source image to facilitate the subsequent image fusion and segmentation process, and the advantages and characteristics of the first source image can be retained. The fusion of the simulated second source image and the second source image is equivalent to the fusion of the real first source image and the second source image, and is also the image fusion process of the same source, without losing any information and affecting the utilization efficiency of multi-source information.

[0047] The training process is further described by taking the first source image as a visible light image and the second source image as a thermal infrared image. Two lines are included in the modal conversion network: a line of input visible light-simulated thermal infrared-reconstructed visible light and a line of input thermal infrared-simulated visible light-reconstructed thermal infrared. Taking the first line as an example, first, the original visible light image is input into the first generator G1 to generate a simulated thermal infrared image, and the simulated thermal infrared image and the real thermal infrared image are input as the authenticity discrimination of the first discriminator D1 and the GAN loss is calculated, then the simulated thermal infrared image is input into the second generator G2 to generate a reconstructed visible light image, and the reconstructed visible light image and the original visible light image are calculated for pixel-level loss. Similarly, in the second line, first, the original thermal infrared image is input into the second generator G2 to generate a simulated visible light image, and the simulated visible light image and the real visible light image are input as the authenticity discrimination of the second discriminator D2 and the GAN loss is calculated, then the simulated visible light image is input into the first generator G1 to generate a reconstructed thermal infrared image, and the reconstructed thermal infrared image and the original thermal infrared image are calculated for pixel-level loss. At the same time, the intermediate features of the first generator G1 called twice in the calculation process are loss-constrained in the form of a twin network, and the intermediate features of the second generator G2 called twice are loss-constrained in the form of a twin network.

[0048] S3: taking the second source image and the simulated second source image as inputs, obtaining a segmentation result by using a segmentation model, and completing target detection.

[0049] The segmentation model of the embodiment is obtained by training a dual-channel segmentation network, and the dual-channel segmentation network is a conditional generative adversarial network cGAN. Figure 3As shown, the dual-channel segmentation network includes an input layer, a convolutional layer, a third generator G3 and a third discriminator D3. The input layer is connected to the convolutional layer and the third discriminator D3, respectively, the convolutional layer is connected to the third generator G3, and the third generator G3 is connected to the third discriminator D3.

[0050] Before S3, the target detection method of the embodiment further includes a step of training the dual-channel segmentation network to obtain a segmentation model, which can include:

[0051] obtaining a second training sample set; the second training sample set includes a plurality of training combinations each composed of a training simulation second source image and a training second source image;

[0052] inputting the training combination to the convolutional layer through the input layer;

[0053] extracting features from the training simulation second source image and the training second source image respectively using the convolutional layer to obtain a first feature spectrum and a second feature spectrum; and merging the first feature spectrum and the second feature spectrum to obtain a merged feature spectrum;

[0054] using the third generator G3 to obtain a segmentation result with the merged feature spectrum as input;

[0055] using the third discriminator D3 to adjust the parameters of the third generator G3 by taking the segmentation result, the training simulation second source image and the training second source image as False discrimination, and taking the reference result, the training simulation second source image and the training second source image as True discrimination, to obtain a trained third generator G-seg; the reference result is a reference segmentation true value, which is a preset value.

[0056] determining whether a second iteration termination condition is reached; the second iteration termination condition can be a preset number of iterations.

[0057] if yes, taking the input layer, the convolutional layer and the trained third generator G-seg as the segmentation model;

[0058] if no, taking the trained third generator G-seg as the third generator G3 in the next iteration, and returning to the step of inputting the training combination to the convolutional layer through the input layer.

[0059] In the training process of the segmentation model, instead of using the method of taking the segmentation result as False discrimination and taking the reference result as True discrimination, the embodiment uses the method of taking the segmentation result, the training simulation second source image and the training second source image as False discrimination, and taking the reference result, the training simulation second source image and the training second source image as True discrimination. The segmentation model trained by this method has a further improved segmentation accuracy.

[0060] Taking the second source image as a thermal infrared image as an example, the training process is further described as follows: the feature spectrum of the simulated thermal infrared image and the real thermal infrared image after several times of convolution in the dual-channel segmentation network is merged, and then the segmentation result is obtained through the third generator G3. Then, the segmentation result, together with the simulated thermal infrared image and the real thermal infrared image, is input into the third discriminator D3 of the cGAN as a False judgment, and the reference result, together with the simulated thermal infrared image and the real thermal infrared image, is input into the third discriminator D3 of the cGAN as a True judgment.

[0061] Specifically, after obtaining the segmentation model, S3 can include:

[0062] The second source image and the simulated second source image are input into the convolution layer through the input layer.

[0063] The convolution layer is used to extract features of the second source image to obtain the feature spectrum of the second source image, and to extract features of the simulated second source image to obtain the feature spectrum of the simulated second source image. The feature spectrum of the second source image and the feature spectrum of the simulated second source image are merged to obtain the merged feature spectrum. Thus, the fusion of the second source image and the simulated second source image is realized, which is equivalent to completing the image fusion process of the first source image and the second source image in the same source image fusion manner, so as to avoid information loss and improve the utilization efficiency of multi-source information.

[0064] The third generator G-seg trained is used to obtain the segmentation result by taking the merged feature spectrum as input. The segmentation result can be a binary image, in which the pixels where the target is located are represented by 1, and the pixels where the background is located are represented by 0.

[0065] The embodiment is applicable to various industrial fields requiring multi-source sensor fusion for scene recognition and analysis, and can be widely applied to target detection, recognition, tracking and other mode application fields under multi-source information, and more specifically can be applied to pedestrian segmentation.

[0066] The embodiment includes two modules, a modal conversion network based on a recurrent generative adversarial network and a dual-channel segmentation network based on a conditional generative adversarial network. After training, only the first generator G1 of the modal conversion network and the input layer, the convolution layer and the third generator G-seg of the dual-channel segmentation network need to be retained in the actual application and deployment stage. The first source image is input into the first generator G1 to generate a simulated second source image. The simulated second source image and the second source image constitute the dual input of the input layer, the convolution layer and the third generator G-seg. Therefore, only the forward propagation calculation of the first generator G1 and the input layer, the convolution layer and the third generator G-seg needs to be run once in the entire actual running stage, so as to obtain the segmentation result.

[0067] The embodiment proposes a multi-source image target detection method based on modal conversion, converts a first source image into a simulated second source image and retains the frequency domain characteristics of the first source image, takes the second source image and the simulated second source image as inputs of a conditional GAN segmentation model at the same time to realize the complementation of multi-source information, and then realizes accurate target detection, thereby providing strong support for the application field of multi-source information target identification and segmentation with similar principles. The simulated second source image converted from the first source image and the second source image collected by the sensor are taken as double inputs of the segmentation model to realize the complementation of double-source images when used for target detection, and improve the detection accuracy.

[0068] Embodiment 2

[0069] The embodiment of the present application can divide the functional modules according to the above-mentioned method examples. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The integrated module can be realized in the form of hardware or in the form of a software functional module. It should be noted that the division of modules in the embodiment of the present application is illustrative, and is only a logical functional division. In actual implementation, there can be another division method.

[0070] In the case of dividing each functional module according to each function, Figure 4 A structure schematic diagram of a target detection system based on multi-source image fusion provided by the embodiment of the present application is shown. As shown in the figure, Figure 4 The target detection system comprises:

[0071] An acquisition module M1 is configured to acquire a first source image and a second source image obtained by photographing a same target; the first source image and the second source image are obtained by photographing the target by using different sensors;

[0072] A conversion module M2 is configured to convert the first source image by using a modal conversion model to obtain a simulated second source image;

[0073] A segmentation module M3 is configured to take the second source image and the simulated second source image as inputs, and obtain a segmentation result by using a segmentation model to complete target detection.

[0074] All related contents of each step involved in the above method embodiment can be cited to the functional description of the corresponding functional module, and will not be repeated here.

[0075] Embodiment 3

[0076] The embodiment is used to provide a target detection device based on multi-source image fusion, comprising:

[0077] A processor; and

[0078] Memory, in which computer-readable program instructions are stored.

[0079] The target detection method described above is executed when the computer-readable program instructions are run by the processor.

[0080] Example 4:

[0081] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the target detection method described above.

[0082] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, the disclosure, and the appended claims in carrying out the claimed invention. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0083] Although the invention has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made therein without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely exemplary descriptions of the invention as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if such modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include such modifications and modifications.

Claims

1. A target detection method based on multi-source image fusion, characterized in that, The method comprises: obtaining a first source image and a second source image obtained by photographing the same target; the first source image and the second source image are obtained by photographing the target using different sensors; performing modal conversion on the first source image using a modal conversion model to obtain a simulated second source image; using the second source image and the simulated second source image as input, obtaining a segmentation result using a segmentation model to complete target detection; wherein, the segmentation model is obtained by training a dual-channel segmentation network; the dual-channel segmentation network is a conditional generative adversarial network; the dual-channel segmentation network comprises an input layer, a convolution layer, a third generator and a third discriminator; the input layer is connected to the convolution layer and the third discriminator respectively; the convolution layer is connected to the third generator; the third generator is connected to the third discriminator; Before using the second source image and the simulated second source image as input to obtain a segmentation result using a segmentation model, the target detection method further comprises a step of training the dual-channel segmentation network to obtain a segmentation model, specifically comprising: obtaining a second training sample set; the second training sample set comprises a plurality of training combinations composed of training simulated second source images and training second source images; inputting the training combination to the convolution layer through the input layer; using the convolution layer to extract features from the training simulated second source image and the training second source image respectively to obtain a first feature spectrum and a second feature spectrum; and merging the first feature spectrum and the second feature spectrum to obtain a merged feature spectrum; using the merged feature spectrum as input, using the third generator to obtain a segmentation result; using the third discriminator to adjust the parameters of the third generator by taking the segmentation result, the training simulated second source image and the training second source image as False discrimination, and taking the reference result, the training simulated second source image and the training second source image as True discrimination, to obtain a trained third generator; determine whether a second iteration termination condition is reached; if yes, take the input layer, the convolution layer and the trained third generator as the segmentation model; if no, take the trained third generator as the third generator in the next iteration, and return to the step of inputting the training combination to the convolution layer through the input layer.

2. The object detection method of claim 1, wherein, The first source image is a visible light image, and the second source image is a thermal infrared image. Alternatively, the first source image is a thermal infrared image, and the second source image is a visible light image.

3. The object detection method of claim 1, wherein, The method comprises: photographing the target at a first angle using a first source sensor to obtain the first source image; and photographing the target at a second angle using a second source sensor to obtain the second source image.

4. The object detection method of claim 1, wherein, The modal conversion model is obtained by training a modal conversion network; the modal conversion network is a recurrent generative adversarial network.

5. The object detection method of claim 4, wherein, The modal conversion network comprises a first generator, a second generator, a first discriminator and a second discriminator; the first generator is connected with the first discriminator and the second generator respectively; the second generator is further connected with the second discriminator; Before modal conversion of the first source image is performed by using the modal conversion model to obtain a simulated second source image, the target detection method further comprises the step of training the modal conversion network to obtain a modal conversion model, specifically comprising: obtaining a first training sample set; the first training sample set comprises a plurality of training first source images and a plurality of training second source images; for each of the training first source images, modal conversion of the training first source image is performed by using the first generator to obtain a training simulated second source image; modal conversion of the training simulated second source image is performed by using the second generator to obtain a training reconstructed first source image, and a first pixel-level loss between the training reconstructed first source image and the training first source image is calculated; a first GAN loss is obtained by using the first discriminator according to the training simulated second source image and any of the training second source images; for each of the training second source images, modal conversion of the training second source image is performed by using the second generator to obtain a training simulated first source image; modal conversion of the training simulated first source image is performed by using the first generator to obtain a training reconstructed second source image, and a second pixel-level loss between the training reconstructed second source image and the training second source image is calculated; a second GAN loss is obtained by using the second discriminator according to the training simulated first source image and any of the training first source images; a first twin semantic loss between intermediate features of the first generator when the training simulated second source image is converted from the training first source image and intermediate features of the first generator when the training reconstructed second source image is converted from the training simulated first source image is calculated, and a second twin semantic loss between intermediate features of the second generator when the training simulated first source image is converted from the training second source image and intermediate features of the second generator when the training reconstructed first source image is converted from the training simulated second source image is calculated; a loss value is calculated according to the first pixel-level loss, the first GAN loss, the second pixel-level loss, the second GAN loss, the first twin semantic loss and the second twin semantic loss, and whether a first iteration end condition is reached is determined according to the loss value; if yes, the iteration is ended, and the first generator used in the current iteration is taken as the modal conversion model; if no, the iteration is continued, parameters of the first generator, the second generator, the first discriminator and the second discriminator are adjusted according to the loss value, and the step of "for each of the training first source images, modal conversion of the training first source image is performed by using the first generator" is returned.

6. A target detection system based on multi-source image fusion, characterized in that, comprises: The acquisition module is configured to acquire a first source image and a second source image obtained by photographing a same target; The first source image and the second source image are obtained by photographing the target using different sensors; The conversion module is configured to perform modal conversion on the first source image using a modal conversion model to obtain a simulated second source image; The segmentation module is configured to take the second source image and the simulated second source image as input, and obtain a segmentation result using a segmentation model to complete target detection, wherein, The segmentation model is obtained by training a dual-channel segmentation network, and the dual-channel segmentation network is a conditional generative adversarial network, The dual-channel segmentation network includes an input layer, a convolution layer, a third generator, and a third discriminator, the input layer is connected to the convolution layer and the third discriminator respectively, the convolution layer is connected to the third generator, and the third generator is connected to the third discriminator, Before taking the second source image and the simulated second source image as input, and obtaining a segmentation result using a segmentation model, the segmentation module further trains the dual-channel segmentation network to obtain a segmentation model, specifically including the following steps: obtain a second training sample set, the second training sample set includes a plurality of training combinations composed of training simulated second source images and training second source images, input the training combination to the convolution layer through the input layer, extract features from the training simulated second source images and the training second source images using the convolution layer to obtain first feature spectrum and second feature spectrum, and merge the first feature spectrum and the second feature spectrum to obtain a merged feature spectrum, take the merged feature spectrum as input, and obtain a segmentation result using the third generator, take the segmentation result, the training simulated second source image, and the training second source image as False discrimination, take the reference result, the training simulated second source image, and the training second source image as True discrimination, adjust the parameters of the third generator using the third discriminator to obtain a trained third generator, determine whether a second iteration termination condition is reached, if yes, take the input layer, the convolution layer, and the trained third generator as a segmentation model, if no, take the trained third generator as the third generator in the next iteration, and return to the step of inputting the training combination to the convolution layer through the input layer.

7. A target detection device based on multi-source image fusion, characterized in that, comprise: a processor; and a memory storing computer readable program instructions, wherein the computer readable program instructions, when executed by the processor, perform the target detection method of any one of claims 1-5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the steps of the target detection method of any one of claims 1-5. The program is executed by the processor to implement the steps of the target detection method of any one of claims 1-5.

Images