CN115223154A - Target detection system and detection method thereof - Google Patents
Target detection system and detection method thereof Download PDFInfo
- Publication number
- CN115223154A CN115223154A CN202210849926.2A CN202210849926A CN115223154A CN 115223154 A CN115223154 A CN 115223154A CN 202210849926 A CN202210849926 A CN 202210849926A CN 115223154 A CN115223154 A CN 115223154A
- Authority
- CN
- China
- Prior art keywords
- image
- images
- group
- space
- resolution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 55
- 230000011218 segmentation Effects 0.000 claims abstract description 34
- 238000012545 processing Methods 0.000 claims abstract description 24
- 230000000903 blocking effect Effects 0.000 claims abstract description 14
- 238000011084 recovery Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 38
- 238000000034 method Methods 0.000 claims description 27
- 238000011176 pooling Methods 0.000 claims description 25
- 238000012549 training Methods 0.000 claims description 21
- 230000004913 activation Effects 0.000 claims description 18
- 238000005070 sampling Methods 0.000 claims description 15
- 238000003709 image segmentation Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000010354 integration Effects 0.000 claims description 6
- 238000000926 separation method Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000004088 simulation Methods 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 7
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000003416 augmentation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000004907 flux Effects 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Abstract
The invention provides a target detection system and a detection method thereof, wherein the target detection method comprises the following steps: s1, performing cross blocking processing on an original space image to obtain a first group of space images and a second group of space images which contain different numbers of image blocks; s2, inputting the image blocks into a semantic segmentation network structure to be segmented one by one, and segmenting the target from the background; s3, integrating the divided image blocks respectively to obtain a first recovery image and a second recovery image; s4, performing edge cutting on the first recovered image and the second recovered image to obtain two images with the same resolution; and S5, performing mask addition on the two images with the same resolution ratio to obtain a final image. The invention uses the convolution neural network to carry out double threshold detection on the coarse contour characteristic and the gray value characteristic of the target, realizes the effective detection of the dark and weak space target with low signal-to-noise ratio and realizes the target segmentation with the signal-to-noise ratio lower than 5.
Description
Technical Field
The invention relates to the technical field of image detection, in particular to a target detection system and a detection method thereof.
Background
For space targets, the observation of the space targets by using a ground-based telescope is a very mature technical means. With the increasing demand of various industries for satellite data, more and more satellites are launched into earth orbit, and a large amount of satellite fragments are brought along, so that the space environment is deteriorated. In order to ensure the safety of the space environment, the tracking of the satellite orbit and the detection and avoidance of the space debris become more important.
For small-sized space debris, no detail contour features are imaged; moreover, the reflected sunlight flux is less, and the detection signal-to-noise ratio of the target is very low, which brings great difficulty to image detection. For a spatial target without detail contour features, a traditional target detection method generally uses a gray value as a detection threshold. By setting a detection threshold, extracting the target with the gray value higher than the set threshold in the original image, which can reduce the detection performance of the dark and weak target with low signal-to-noise ratio and no obvious difference from the image background and noise in the image. And if the target gray value is higher than the detection threshold, determining as the target. However, the gray scale value of the low snr target is not significantly different from the sky background, and it is difficult to achieve high efficiency detection.
In view of the above requirements, there are many solutions at home and abroad.
Patent CN107316004A proposes a spatial target identification method based on deep learning. Firstly, a 9-layer deep convolution network model is constructed, then an optimal data augmentation method is found out on the basis of the network, data obtained by a plurality of better augmentation methods are combined, and the optimal combined data is used in the model training and testing processes at the same time to complete space target identification. The method is simple and feasible, but the identification capability of the method is limited, and when the space target is identified, the input data is an image with obvious characteristics and larger size, and the identification effect on the small target is not good.
Patent CN111191583A proposes a spatial target recognition system and method based on convolutional neural network. The method comprises model training and target recognition, wherein the model training comprises inputting positive and negative samples, extracting features by using a sparse convolutional neural network, and training a classifier to form a learning dictionary. The images can be classified through a training network, and the images containing the space target are classified with the images without the space target. Although the method can preliminarily judge whether the image contains the space target, the number of the targets and the specific positions of the targets cannot be further judged at the pixel level.
Patent CN109188462A proposes a method and apparatus for detecting spatial target under complex starry sky background. The method obtains measurement requirements, obtains parameters of the cooperation targets, obtains working data of the star sensor, and designs a relative angle measurement algorithm to obtain a space target detection result. The method can effectively exert the advantages of high precision and high update rate of the star sensor in the field of space detection by applying the performance advantages of the star sensor to the space target detection, but can not accurately detect unknown targets without prior knowledge.
Disclosure of Invention
In view of the foregoing problems, an object of the present invention is to provide a target detection system and a detection method thereof, in which a convolutional neural network is used to perform a dual threshold test on a target coarse contour feature and a gray scale value feature, so as to achieve effective detection of a low signal-to-noise ratio dark and weak space target whose gray scale value is not significantly different from an image background and noise, and achieve segmentation of a target whose signal-to-noise ratio is lower than 5.
In order to achieve the purpose, the invention adopts the following specific technical scheme:
the invention provides a target detection method, which comprises the following steps:
s1, performing cross blocking processing on an original space image, and respectively adjusting the resolution of the original space image to a first set resolution and a second set resolution to obtain a first group of space images and a second group of space images which contain image blocks with different quantities;
s2, inputting image blocks in the first group of space images and the second group of space images into a semantic segmentation network structure to be segmented one by one, so that a target is separated from a background;
s3, integrating the segmented image blocks in the first group of spatial images and the second group of spatial images respectively to obtain a first restored image and a second restored image;
s4, performing edge cutting on the first recovered image and the second recovered image to obtain two images with the same resolution;
and S5, performing mask addition on the two images with the same resolution ratio to obtain a final image.
Preferably, step S2 comprises the following sub-steps:
s21, performing up-sampling operation on image blocks in the first group of space images and the second group of space images one by one;
s22, after the up-sampling operation is finished, performing pooling operation on the divided image blocks;
and S23, after the pooling operation is finished, inputting the pooled image blocks into an activation function of an output layer to obtain image blocks with separated targets and backgrounds.
Preferably, the first set resolution is 2050 × 2050, and the process of adjusting the resolution of the original spatial image to the first set resolution is as follows:
and (3) carrying out zero filling processing on the original space image in the horizontal and vertical directions, and dividing the original space image into 100 small-size image blocks with 205 × 205 resolutions to obtain a first group of space images.
Preferably, the second set resolution is 2255 — 2255, and the process of adjusting the resolution of the original aerial image to the second set resolution is:
and (3) performing zero filling processing on the boundary of the original space image in the horizontal and vertical directions, and dividing the boundary into 121 parts of small-size image blocks with 205 × 205 resolutions to obtain a second group of space images.
Preferably, the semantic segmentation network structure consists of 12 convolutional layers and 3 pooling layers, and an activation function and a BN layer are arranged behind each convolutional layer;
the other activation functions are linear rectification functions except that the activation function set after the last convolution layer is a sigmoid function.
Preferably, the first group of spatial images are integrated to obtain a first restored image, and the resolution of the first restored image is 2050 × 2050;
the second set of aerial images is integrated to produce a second restored image having a resolution of 2255 x 2255.
Preferably, the method for training the simulated image of the semantic segmentation network in step S2 includes the following steps:
inputting image blocks with the resolution of 205 × 205 into a semantic segmentation network;
the image block f (x, y) is preprocessed by adopting the image standardization, and the image standardization formula is as follows:
wherein std (f (x, y)) is a standard deviation calculation function, and min (f (x, y)) is a minimum calculation function;
in the training process of the semantic segmentation network structure, loss calculation is carried out on each pixel point in the image block through a loss function, and pixel-level classification is completed;
the loss function is formulated as follows:
wherein, N is the number of pixels, and y and p are the label and the network output respectively;
after 100 rounds of training, the simulation training of the semantic segmentation network is completed.
The present invention also provides a target detection system, comprising: the image segmentation module comprises an image segmentation module, an image integration module, an image cutting module and an image addition module;
the image blocking module is used for carrying out cross blocking processing on the original space image, and obtaining a first group of space images and a second group of space images containing image blocks with different quantities after respectively adjusting the resolution of the original space image to a first set resolution and a second set resolution;
the image segmentation module is used for inputting image blocks in the first group of space images and the second group of space images into a semantic segmentation network structure to be segmented one by one, and segmenting the target from the background;
the image integration module is used for respectively integrating the image blocks after being divided in the first group of spatial images and the second group of spatial images to obtain a first recovery image and a second recovery image;
the image cutting module is used for cutting the edges of the first recovered image and the second recovered image to obtain two images with the same resolution;
the image adding module is used for carrying out mask addition on the two images with the same resolution ratio to obtain a final image.
Preferably, the image segmentation module comprises: the device comprises an up-sampling unit, a pooling unit and an image block separating unit;
the up-sampling unit is used for performing up-sampling operation on the image blocks in the first group of spatial images and the second group of spatial images one by one;
the pooling unit performs pooling operation on the divided image blocks;
and the image block separation unit is used for inputting the pooled image blocks into the activation function of the output layer to obtain image blocks with separated targets and backgrounds.
Preferably, the image blocking module includes an image zero padding processing unit;
the zero filling processing unit is used for dividing the original space image into 100 small-size image blocks with 205 × 205 resolutions, and then a first group of space images are obtained;
the zero padding processing unit is further configured to divide the original spatial image into 121 small-size image blocks with a resolution of 205 × 205, that is, to obtain a second set of spatial images.
Compared with the traditional space target detection method, the method mainly has the following advantages:
(1) The effects of recognizing and detecting the dim and weak targets are better. The method is different from the traditional energy-based detection method, and the deep learning method can better combine the characteristics of the shape outline, the gray level and the like of the target so as to realize good segmentation and detection effects on the dark and weak target.
(2) The segmented image is clear, and the tracking of the tracker is facilitated. The segmented image after forward propagation by using the deep learning network has no clutter and only a space target and a background, and the processed image target is clear and definite, thereby being beneficial to further tracking the space target by a subsequent target tracker.
(3) Cloud cover, vignetting and other detector equipment have little interference. The invention obtains the characteristics capable of overcoming the interference by training the proper neural network. By adding the data with interference such as cloud layers, vignetting and the like into the training set, the adaptation of the network to different environments can be enhanced without additionally writing a denoising program.
Drawings
Fig. 1 is a schematic flow chart of a target detection method according to an embodiment of the present invention.
Fig. 2 is a schematic diagram illustrating an overall step of spatial target detection in the target detection method according to an embodiment of the present invention.
Fig. 3 is a schematic diagram of image cross blocking processing of the object detection method according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of a semantic segmentation network structure of an object detection method according to an embodiment of the present invention.
Detailed Description
Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the following description, the same reference numerals are used for the same blocks. In the case of the same reference numerals, their names and functions are also the same. Therefore, detailed description thereof will not be repeated.
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the invention.
Fig. 1 illustrates a target detection method provided according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of the overall steps of spatial target detection in the target detection method according to the embodiment of the present invention.
As shown in fig. 1 and fig. 2, the target detection method provided in the embodiment of the present invention includes the following steps:
s1, respectively carrying out cross blocking processing on the original space images, and respectively adjusting the resolutions of the original space images to a first set resolution and a second set resolution to obtain a first group of space images and a second group of space images which contain image blocks with different quantities.
Step S1 is to construct a suitable size for subsequent transmission into the segmentation network and to eliminate the detection influence of the target on the block boundary.
The preset resolution of the original spatial target image is 2048 × 2048.
Fig. 3 is a schematic diagram of image cross blocking processing of the target detection method according to the embodiment of the present invention.
As shown in fig. 3, the process of adjusting the resolution of the original spatial image to the first set resolution is as follows:
the scale of the original space image is modified to 2050 × 2050, that is, the original space image is subjected to zero filling processing in the horizontal and vertical directions, and then the image is divided into 100 small-size image blocks with 205 × 205 resolution, that is, a first group of space images, and the first group of space images are input to the semantic segmentation network in step S2 for segmentation operation.
The process of adjusting the resolution of the original spatial image to the second set resolution is as follows:
the scale of the original spatial image is modified to 2255-2255, zero padding is also performed at the boundary of the original spatial image, and then the image is divided into 121 small-size image blocks with 205-205 resolution, i.e. a second set of spatial images, and the second set of spatial images is input to the semantic segmentation network in step S2 for segmentation.
The boundary of the first group of images is just the middle line of the second group of images, so that the influence of the target falling on the boundary of the image block is eliminated.
And S2, inputting image blocks in the first group of space images and the second group of space images into a semantic segmentation network to be segmented one by one, and segmenting the target from the background.
Fig. 4 is a schematic diagram of a semantic segmentation network structure of an object detection method according to an embodiment of the present invention.
As shown in fig. 4, the semantic segmentation network structure mainly comprises 12 convolutional layers and 3 pooling layers, and an activation function and a BN (batch normalization) layer are arranged behind each convolutional layer. Except that the activation function set after the last convolution layer is a sigmoid (S-shaped growth curve) function, the other activation functions are equal to ReLU (linear rectification function), and bilinear interpolation upsampling operation is carried out on the image at the beginning of the network so as to increase the size of the image block and the characteristics of the dark and weak targets.
And S21, performing up-sampling operation on the image blocks in the first group of spatial images and the second group of spatial images one by one.
The upsampling operation provided by the invention is specifically processed by bilinear interpolation, and the specific formula of the bilinear interpolation is as follows:
wherein Q 11 、Q 12 、Q 21 、Q 22 Are respectively four adjacent pixel points, x 1 、x 2 、y 1 、y 2 And x and y are coordinates of the point to be interpolated.
The bilinear interpolation processing can ensure the accuracy of the segmentation result and reduce the parameters at the same time.
The interpolation is to linearly utilize the neighboring pixel values, so as to achieve the purpose of increasing the resolution of the image. Convolution is an important structure for processing a feature map, and the specific content is as follows:
where f (x, y) is the feature diagram to be convolved, w is the size (2a +1,2b + 1) as the convolution kernel, and s and t are the convolution kernels at different positions.
By iteratively training the parameters of the convolution kernel, the convolution parameters that can extract the target features are continuously optimized.
And S22, after the up-sampling operation is finished, performing pooling operation on the divided image blocks.
The pooling operation adopted by the invention is maximum pooling, namely the maximum value of the reserved area is used as lower-layer feature input, the pooling step length is 2, and the resolution of the feature map is reduced by half after each pooling. Pooling is used as an important operation in the convolutional neural network, and can be matched with convolution to extract image characteristics, so that the characteristic invariance is ensured. And the pooling can simplify the network complexity and reduce the amount of computation. Further, pooling may prevent overfitting of the network to some extent.
And S23, after the pooling operation is finished, inputting the pooled image blocks into an activation function of an output layer to obtain image blocks with separated targets and backgrounds.
The invention simultaneously uses sigmoid (S-shaped growth curve) and ReLU (linear rectification function) activation functions, and the formulas are respectively
The activation function in the network can increase the nonlinearity of the network and optimize the fitting effect of the network. The activation function of the output layer is used for pixel classification, which can better separate the target from the background.
The invention also provides a method for training the simulated image of the semantic segmentation network, which specifically comprises the following steps:
the training data set is the data of the constructed simulated image training set, the resolution of the image block is 205 × 205, and the number of image channels is 1.
Preprocessing the image block by adopting the standardization of the image, wherein the image standardization formula is as follows:
wherein std (f (x, y)) is a standard deviation calculation function, and min (f (x, y)) is a minimum calculation function;
the optimizer selected in the training process of the semantic segmentation network structure is SGD (random gradient descent), and the initial learning rate is 1 × 10 -4 The learning rate decays to 5 × 10 -7 More accurate network weight updates are achieved through a decaying learning rate. The loss function selects the cross entropy, which is shown in equation (5):
wherein, N is the number of pixels, and y and p are the label and the network output respectively.
The loss function can calculate loss of each pixel point in the image block, and pixel-level classification is completed.
Because the resolution of the image input into the semantic segmentation network is very high and the video memory of a GPU (graphics processing unit) is limited, the number of image blocks input into the semantic segmentation network is 1, and the training is performed for 100 rounds.
And S3, integrating the image blocks after being divided in the first group of space images and the second group of space images respectively to obtain a first restored image and a second restored image.
And integrating the first group of spatial images to obtain a first restored image, wherein the resolution of the first restored image is 2050 by 2050.
The second set of aerial images are integrated to produce a second restored image having a resolution of 2255 x 2255.
And S4, performing edge cutting on the first recovered image and the second recovered image to obtain two images with the same resolution.
The resolution of two images of the same resolution is 2048 × 2048.
And S5, performing mask addition on the two images with the same resolution ratio to obtain a final image.
The present invention also provides a target detection system, comprising: the image segmentation module comprises an image segmentation module, an image integration module, an image cutting module and an image addition module;
the image blocking module is used for performing cross blocking processing on the original space image, and obtaining a first group of space images and a second group of space images containing image blocks with different quantities after respectively adjusting the resolution of the original space image to a first set resolution and a second set resolution;
the image segmentation module is used for inputting image blocks in the first group of space images and the second group of space images into a semantic segmentation network structure to be segmented one by one, and segmenting the target from the background;
the image segmentation module comprises: the device comprises an up-sampling unit, a pooling unit and an image block separating unit;
the up-sampling unit is used for performing up-sampling operation on the image blocks in the first group of spatial images and the second group of spatial images one by one;
the pooling unit performs pooling operation on the divided image blocks;
and the image block separation unit is used for inputting the pooled image blocks into the activation function of the output layer to obtain image blocks with separated targets and backgrounds.
The image integration module is used for respectively integrating the image blocks after being divided in the first group of space images and the second group of space images to obtain a first recovery image and a second recovery image;
the image cutting module is used for cutting the edges of the first recovered image and the second recovered image to obtain two images with the same resolution;
the image adding module is used for performing mask addition on the two images with the same resolution ratio to obtain a final image.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
The above embodiments of the present invention should not be construed as limiting the scope of the present invention. Any other corresponding changes and modifications made according to the technical idea of the present invention should be included in the protection scope of the claims of the present invention.
Claims (10)
1. A method of target detection, comprising the steps of:
s1, performing cross blocking processing on an original space image, and respectively adjusting the resolution of the original space image to a first set resolution and a second set resolution to obtain a first group of space images and a second group of space images which contain different numbers of image blocks;
s2, inputting image blocks in the first group of space images and the second group of space images into a semantic segmentation network structure to be segmented one by one so as to separate a target from a background;
s3, integrating the image blocks after being divided in the first group of space images and the second group of space images respectively to obtain a first recovery image and a second recovery image;
s4, performing edge cutting on the first restored image and the second restored image to obtain two images with the same resolution;
and S5, performing mask addition on the two images with the same resolution ratio to obtain a final image.
2. The object detection method according to claim 1, characterized in that said step S2 comprises the following sub-steps:
s21, performing up-sampling operation on image blocks in the first group of spatial images and the second group of spatial images one by one;
s22, after the up-sampling operation is finished, performing pooling operation on the segmented image blocks;
and S23, after the pooling operation is finished, inputting the pooled image blocks into an activation function of an output layer to obtain image blocks with separated targets and backgrounds.
3. The object detection method according to claim 2, wherein the first set resolution is 2050 × 2050, and the step of adjusting the resolution of the original spatial image to the first set resolution is:
and (3) carrying out zero filling processing on the original space images in the horizontal and vertical directions, and dividing the original space images into 100 small-size image blocks with 205 × 205 resolutions to obtain the first group of space images.
4. The object detection method according to claim 3, wherein the second set resolution is 2255 x 2255, and the process of adjusting the resolution of the original spatial image to the second set resolution is:
and (3) performing zero filling processing on the boundary of the original space image in the horizontal and vertical directions, and dividing the boundary into 121 parts of small-size image blocks with 205 × 205 resolutions to obtain the second group of space images.
5. The object detection method of claim 4, wherein the semantic segmentation network structure is composed of 12 convolutional layers and 3 pooling layers, and an activation function and a BN layer are arranged after each convolutional layer;
the other activation functions are linear rectification functions except that the activation function set after the last convolution layer is a sigmoid function.
6. The object detection method according to claim 5,
integrating the first group of spatial images to obtain a first restored image, wherein the resolution of the first restored image is 2050 × 2050;
the second set of aerial images is integrated to produce a second restored image having a resolution of 2255 x 2255.
7. The object detection method according to claim 1, wherein the simulated image training method of the semantic segmentation network in the step S2 comprises the following steps:
inputting image blocks with the resolution of 205 × 205 into a semantic segmentation network;
preprocessing the image block f (x, y) by adopting image standardization, wherein an image standardization formula is as follows:
wherein std (f (x, y)) is a standard deviation calculation function, and min (f (x, y)) is a minimum calculation function;
performing loss calculation on each pixel point in the image block through a loss function in the training process of the semantic segmentation network structure to complete pixel-level classification;
the loss function is formulated as follows:
wherein, N is the number of pixels, and y and p are the label and the network output respectively;
and after 100 rounds of training, completing simulation training of the semantic segmentation network.
8. An object detection system, comprising: the image segmentation module comprises an image segmentation module, an image integration module, an image cutting module and an image addition module;
the image blocking module is used for performing cross blocking processing on an original space image, and obtaining a first group of space images and a second group of space images containing image blocks with different quantities after respectively adjusting the resolution of the original space image to a first set resolution and a second set resolution;
the image segmentation module is used for inputting image blocks in the first group of spatial images and the second group of spatial images into a semantic segmentation network structure to perform segmentation processing one by one, and segmenting a target from a background;
the image integration module is used for respectively integrating the image blocks after being divided in the first group of spatial images and the second group of spatial images to obtain a first recovery image and a second recovery image;
the image cutting module is used for performing edge cutting on the first recovery image and the second recovery image to obtain two images with the same resolution;
and the image adding module is used for performing mask addition on the two images with the same resolution ratio to obtain a final image.
9. The object detection system of claim 8, wherein the image segmentation module comprises: the device comprises an up-sampling unit, a pooling unit and an image block separating unit;
the up-sampling unit is used for performing up-sampling operation on image blocks in the first group of spatial images and the second group of spatial images one by one;
the pooling unit performs pooling operation on the segmented image blocks;
and the image block separation unit is used for inputting the pooled image blocks into the activation function of the output layer to obtain the image blocks with separated targets and backgrounds.
10. The object detection system of claim 9, wherein the image blocking module comprises an image zero padding processing unit;
the zero padding processing unit is used for dividing the original space image into 100 small-size image blocks with 205 × 205 resolutions, namely obtaining the first group of space images;
the zero padding processing unit is further configured to divide the original spatial image into 121 small-size image blocks with 205 × 205 resolutions, i.e., obtain the second group of spatial images.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210849926.2A CN115223154A (en) | 2022-07-20 | 2022-07-20 | Target detection system and detection method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210849926.2A CN115223154A (en) | 2022-07-20 | 2022-07-20 | Target detection system and detection method thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115223154A true CN115223154A (en) | 2022-10-21 |
Family
ID=83611107
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210849926.2A Pending CN115223154A (en) | 2022-07-20 | 2022-07-20 | Target detection system and detection method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115223154A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921196A (en) * | 2018-06-01 | 2018-11-30 | 南京邮电大学 | A kind of semantic segmentation method for improving full convolutional neural networks |
| CN112991354A (en) * | 2021-03-11 | 2021-06-18 | 东北大学 | High-resolution remote sensing image semantic segmentation method based on deep learning |
| US20210279887A1 (en) * | 2020-03-04 | 2021-09-09 | Alibaba Group Holding Limited | Method and system for performing image segmentation |
-
2022
- 2022-07-20 CN CN202210849926.2A patent/CN115223154A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921196A (en) * | 2018-06-01 | 2018-11-30 | 南京邮电大学 | A kind of semantic segmentation method for improving full convolutional neural networks |
| US20210279887A1 (en) * | 2020-03-04 | 2021-09-09 | Alibaba Group Holding Limited | Method and system for performing image segmentation |
| CN112991354A (en) * | 2021-03-11 | 2021-06-18 | 东北大学 | High-resolution remote sensing image semantic segmentation method based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109584248B (en) | Infrared target instance segmentation method based on feature fusion and dense connection network | |
| CN110119728B (en) | Remote sensing image cloud detection method based on multi-scale fusion semantic segmentation network | |
| CN108921799B (en) | Remote sensing image thin cloud removing method based on multi-scale collaborative learning convolutional neural network | |
| CN109934200B (en) | RGB color remote sensing image cloud detection method and system based on improved M-Net | |
| CN111738124B (en) | Remote sensing image cloud detection method based on Gabor transformation and attention | |
| CN111862126B (en) | Non-cooperative target relative pose estimation method combining deep learning and geometric algorithm | |
| CN108647655B (en) | Low-altitude aerial image power line foreign matter detection method based on light convolutional neural network | |
| CN112861729B (en) | Real-time depth completion method based on pseudo-depth map guidance | |
| CN107563433B (en) | Infrared small target detection method based on convolutional neural network | |
| CN112184577B (en) | Single image defogging method based on multiscale self-attention generation countermeasure network | |
| CN111507271A (en) | Airborne photoelectric video target intelligent detection and identification method | |
| CN107808138B (en) | Communication signal identification method based on FasterR-CNN | |
| CN106295613A (en) | A kind of unmanned plane target localization method and system | |
| CN111008664B (en) | Hyperspectral sea ice detection method based on space-spectrum combined characteristics | |
| CN113838064B (en) | Cloud removal method based on branch GAN using multi-temporal remote sensing data | |
| CN113610905B (en) | Deep learning remote sensing image registration method based on sub-image matching and application | |
| CN114022408A (en) | Remote sensing image cloud detection method based on multi-scale convolution neural network | |
| CN111008975A (en) | Mixed pixel unmixing method and system for space artificial target linear model | |
| CN110647977B (en) | Method for optimizing Tiny-YOLO network for detecting ship target on satellite | |
| CN115995042A (en) | Video SAR moving target detection method and device | |
| CN115861756A (en) | Earth background small target identification method based on cascade combination network | |
| CN115631427A (en) | Multi-scene ship detection and segmentation method based on mixed attention | |
| CN113486819A (en) | Ship target detection method based on YOLOv4 algorithm | |
| CN117351333A (en) | Quick star image extraction method of star sensor | |
| CN116385281A (en) | Remote sensing image denoising method based on real noise model and generated countermeasure network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |