CN114241285B - Ship rapid detection method based on knowledge distillation and semi-supervised learning - Google Patents
Ship rapid detection method based on knowledge distillation and semi-supervised learning Download PDFInfo
- Publication number
- CN114241285B CN114241285B CN202111410408.2A CN202111410408A CN114241285B CN 114241285 B CN114241285 B CN 114241285B CN 202111410408 A CN202111410408 A CN 202111410408A CN 114241285 B CN114241285 B CN 114241285B
- Authority
- CN
- China
- Prior art keywords
- sample
- ship
- data
- network
- student network
- 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.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 46
- 238000013140 knowledge distillation Methods 0.000 title claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 35
- 230000006870 function Effects 0.000 claims description 40
- 238000002372 labelling Methods 0.000 claims description 25
- 238000000034 method Methods 0.000 claims description 21
- 238000012544 monitoring process Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000002776 aggregation Effects 0.000 claims description 3
- 238000004220 aggregation Methods 0.000 claims description 3
- 230000002708 enhancing effect Effects 0.000 abstract description 2
- 238000010276 construction Methods 0.000 description 5
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000011478 gradient descent method Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a ship rapid detection method based on knowledge distillation and semi-supervised learning, which comprises the following steps: 1) Constructing a ship data set by utilizing the existing ship data and the ship data acquired from the sea area to be deployed, marking ship targets on part of the data in the data set, generating training samples according to priori frame sliding and enhancing the data; 2) Constructing a teacher network, pre-training by using a marked sample, and then performing semi-supervised training by combining a non-marked sample to realize ship detection; 3) Constructing a lightweight student network, and guiding the student network to realize knowledge distillation on the marked sample and the unmarked sample by using a trained teacher network; 4) And converting the trained student network model into ONNX format, and further optimizing and deploying the student network by adopting OpenVINO to realize rapid ship detection. According to the invention, the knowledge learned by the teacher network with high precision and large capacity is transferred to the lightweight student network, so that the detection precision is ensured, and meanwhile, the rapid ship detection is realized.
Description
Technical Field
The invention relates to the technical field of ship detection, in particular to a ship rapid detection method based on knowledge distillation and semi-supervised learning.
Background
Currently, offshore economy is developed vigorously, real-time monitoring of the offshore environment is helpful for timely finding out events such as blockage, smuggling and the like, and management level is improved. However, a large amount of monitoring image data is difficult to effectively process in a manual mode, so that rapid ship detection is realized, unknown targets can be early-warned in real time, information classification of ships is improved, an informationized ship management system is constructed, and the method has important theoretical significance and application value.
The purpose of ship detection is to obtain the position and class of the ship in the input image. At present, ship detection methods mainly comprise two main categories: a ship detection method based on background modeling and a ship detection method based on deep learning. The ship detection method based on background modeling mainly comprises the steps of modeling a background, and then rapidly extracting a foreground target by utilizing background subtraction. However, such methods tend to misidentify stationary vessels as background, and a large number of false positive targets are generated in dynamically changing offshore scenarios. Furthermore, such methods lack further class discrimination of detected targets. The ship detection method based on deep learning mainly utilizes a neural network to extract target characteristics, and carries out fine granularity classification and position optimization on targets. However, such methods are prone to overfitting risks due to the large number of parameters, in the case of small data set sizes. And a large number of parameters lead the detection model to be dependent on special equipment such as a display card for accelerating operation, and real-time ship detection is difficult to realize in general equipment.
Disclosure of Invention
The invention aims to overcome the defects and shortcomings of the prior art, and provides a ship rapid detection method based on knowledge distillation and semi-supervised learning, wherein a teacher network with high detection precision is trained by combining a marked sample and a non-marked sample in a semi-supervised learning mode, then a light-weight student network is trained by knowledge distillation, high detection precision is maintained, and rapid detection of a marine ship is realized on a target candidate region extracted from an image.
In order to achieve the above purpose, the technical scheme provided by the invention is as follows: a ship rapid detection method based on knowledge distillation and semi-supervised learning comprises the following steps:
1) Constructing a ship data set by utilizing the existing ship data and the ship data collected by the deployed sea area, marking a ship target on part of the data in the data set, generating a training sample according to priori frame sliding and carrying out data enhancement, wherein the training sample comprises a marked sample and a non-marked sample;
2) Constructing a teacher network, pre-training by using a marked sample, and then performing semi-supervised training by combining a non-marked sample to realize ship detection;
3) Constructing a lightweight student network, and guiding the student network to realize knowledge distillation on the marked sample and the unmarked sample by using a trained teacher network;
4) And converting the trained student network into ONNX format, adopting OpenVINO to further optimize and deploy the student network, and finally realizing rapid ship detection through the optimized student network.
Further, the step 1) includes the steps of:
1.1 Acquiring existing ship data and acquiring data in a sea area to be deployed, constructing a ship data set, marking ship targets on part of the data to form marked data and unmarked data, and acquiring a multi-scale priori frame through clustering calculation of the ship targets in the marked data;
1.2 According to the sliding of the multi-scale prior frame, generating a multi-scale ship sample serving as a training sample, wherein the multi-scale ship sample comprises the following specific operations: sliding in the labeling data to generate a labeling sample, and dividing the labeling sample into a positive sample and a negative sample according to the intersection ratio of the sample and a labeling frame, wherein the positive sample comprises a multi-class ship target, records a class label and a position regression label of the multi-class ship target, and the negative sample is a background sample and slides in the non-labeling data to generate a non-labeling sample;
1.3 Data enhancement for training samples including horizontal flipping, color dithering, light variation, and random clipping.
Further, in step 2), a teacher network is constructed and training is performed, including the following steps:
2.1 Constructing a teacher network with large model capacity as a ship discriminator, and pre-training the network by using a labeling sample;
2.2 Semi-supervised objective function for constructing teacher network
Where D L represents the set of marked samples, D U represents the set of unmarked samples, |D L | represents the number of marked samples, |D U | represents the number of unmarked samples, D represents one sample in the set, [ theta ] t is a trainable parameter of the teacher network,To monitor loss,/>For the consistency regularization term, λ 1 represents the penalty coefficient.
Monitoring lossesThe method comprises the steps of category loss and regression loss, wherein the category loss is calculated by adopting a cross entropy function, and the regression loss is calculated by adopting a smoothl 1 function; the purpose of the consistency regular term is to enable different data enhancement samples of the same sample to have category consistency and improve the category aggregation degree of target features, and the mathematical form is as follows:
where N represents the number of data enhancements, f c represents the cross entropy function, f t represents the sample prediction function of the teacher network, d (0) represents the original sample, and d (i) represents the ith data enhanced sample of the original sample.
Further, the step 3) includes the steps of:
3.1 A student network with less parameters is built as a lightweight ship discriminator;
3.2 Knowledge distillation objective function for constructing student network
Wherein D L represents a set of marked samples, D U represents a set of unmarked samples, |D L | represents the number of marked samples, |D U | represents the number of unmarked samples, D represents one sample in the set, θ s is a trainable parameter of the student network,To monitor loss,/>For knowledge distillation regularization term, λ 2 represents a penalty coefficient;
Monitoring losses The method comprises the steps of category loss and regression loss, wherein the category loss is calculated by adopting a cross entropy function, and the regression loss is calculated by adopting a smoothl1 function; the purpose of the knowledge distillation regular item is to migrate the knowledge learned by the teacher network to the student network, so that the student network can have the prediction of consistency with the teacher network, and the mathematical form is as follows:
Wherein f c represents a cross entropy function, f l represents a smooth-L1 function, f s represents a sample prediction function of a student network, f t represents a sample prediction function of a teacher network, f T represents a sharpening function, T represents a sharpening coefficient, and the mathematical form of the sharpening function is as follows:
where p represents a one-dimensional input vector, Represents/>, of the j-th component of pTo the power.
Further, the step 4) includes the steps of:
4.1 Converting the trained student network into ONNX format, further optimizing the student network by OpenVINO and deploying;
4.2 The input image is cut into 4 non-overlapping areas, and the 4 areas are monitored circularly through the optimized student network, so that the ship detection speed is further improved.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. According to the technical scheme of combining knowledge distillation and semi-supervised learning, knowledge learned by a teacher network is transferred to a lightweight student network through the consistency regular term and the knowledge distillation regular term, the technical problems of high labeling cost and large model parameter number in a full-supervised learning method are solved, and the technical effect of improving the detection speed while guaranteeing the detection precision is achieved.
2. According to the invention, the candidate region is generated by utilizing the multi-scale prior frame, so that the ship detection problem is converted into the classification and regression problem, the calculation complexity is reduced, and the lightweight network model is adopted to deploy the actual scene.
3. The invention converts the trained network model into ONNX format, can make the network model be repeatedly called in different frames, adopts OpenVINO to deploy and infer the network model, and is beneficial to further accelerating ship detection on CPU.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Fig. 2 is a schematic diagram of the positive sample generated.
Fig. 3 is a graph of accuracy of classification of teacher and student networks.
Fig. 4 is a diagram of a ship detection result achieved by the method of the present invention.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but embodiments of the present invention are not limited thereto.
The experimental platform of this embodiment is python3.6, pytorch1.7.1, computer configuration: CPU model is Intel (R) Core (TM) i9-10900X, memory is 32GB, and display card model is NVIDIA Quadro M6000.
As shown in fig. 1, the method for quickly detecting a ship based on knowledge distillation and semi-supervised learning provided in this embodiment is roughly divided into four stages: the first stage is the construction and pretreatment of a data set, mainly the construction of a ship data set, the calculation of a ship priori frame, the generation of a training sample and the data enhancement; the second stage is the construction and training of a teacher network, mainly constructing a teacher network with large model capacity, and performing pre-training and semi-supervision training; the third stage is the construction and training of the student network, mainly constructing a lightweight student network, and transferring the knowledge of the teacher network to the student network by using knowledge distillation; the fourth stage is the optimization and deployment of the model, mainly converting the trained student network into ONNX format and deploying by OpenVINO. The method comprises the following specific steps:
1) Constructing a ship data set by utilizing the existing ship data and the ship data collected by the deployed sea area, marking a ship target on part of the data in the data set, generating training samples (including marked samples and unmarked samples) according to priori frame sliding and enhancing the data, wherein the method comprises the following steps of:
1.1 Construction and processing of data sets: collecting data in the sea A area, utilizing SEA MARITIME DATASET to disclose a data set, cutting all images into 576 multiplied by 704 pictures according to a camera data format, and constructing a ship data set. And labeling the ship target on part of the data in the sea area A to form labeling data and non-labeling data. The multi-scale prior frames of the ship targets are mainly obtained by clustering calculation on the ship target frames in the labeling data, and the number of the multi-scale prior frames is 5: (24, 32), (39, 69), (68, 141), (139, 297), (302, 689), wherein each a priori box is represented in (height, width).
1.2 Training sample generation: and sliding in the image of the ship data set according to the multi-scale prior frame to generate a multi-scale ship sample serving as a training sample. And distinguishing positive and negative samples according to the intersection ratio of the sample frame and the labeling frame in the labeling data according to the labeling samples generated by sliding the multi-scale prior frame, wherein the intersection ratio is greater than 0.4 and is a positive sample, and the intersection ratio is less than 0.2 and is a negative sample. The positive samples comprise multi-category ship targets, category labels and position regression labels of the ship targets are recorded, and the negative samples are background samples. And generating unlabeled samples in the unlabeled data according to the sliding of the multi-scale prior frames, and storing the unlabeled samples in the same folder. As shown in fig. 2, the positive sample labeling information consists of a sample name, a category label and a position regression vector t. The position regression vector is calculated as follows:
Wherein g x,gy,gw and g h are respectively the central abscissa, central ordinate, width and length of the target labeling frame corresponding to the positive sample, and a x,ay,aw and a h are respectively the central abscissa, central ordinate, width and length of the positive sample sliding frame.
1.3 Data enhancement: data enhancement is carried out on the generated training samples, including horizontal overturning, color dithering, light change and random cutting, so as to improve the diversity of the training samples;
2) The teacher network is constructed, and the high-precision ship detection is realized through pre-training and semi-supervised training, and the method comprises the following steps:
2.1 Pre-training: a teacher network based on Resnet110,110 is built as a ship discriminator, the teacher network is firstly subjected to classification training by using a large-scale classification task data set, and then classification and regression branches of the teacher network are further trained by using a labeling sample. Wherein all of the input training samples are scaled to a size of 32 x 32.
2.2 A semi-supervised objective function of the teacher network is built and optimized by adopting a random gradient descent method:
where D L represents the set of marked samples, D U represents the set of unmarked samples, |D L | represents the number of marked samples, |D U | represents the number of unmarked samples, D represents one sample in the set, [ theta ] t is a trainable parameter of the teacher network, To monitor loss,/>For the consistency regularization term, λ 1 represents the penalty coefficient.
Monitoring lossesThe method comprises the steps of category loss and regression loss, wherein the category loss is calculated by adopting a cross entropy function, and the regression loss is calculated by adopting a smoothl 1 function. The purpose of the consistency regular term is to enable different data enhancement samples of the same sample to have category consistency and improve the category aggregation degree of target features, and the mathematical form is as follows:
where N represents the number of data enhancements, f c represents the cross entropy function, f t represents the sample prediction function of the teacher network, d (0) represents the original sample, and d (i) represents the ith data enhanced sample of the original sample.
3) Constructing a lightweight student network, guiding the student network to realize knowledge distillation on marked samples and unmarked samples by using a trained teacher network, and comprising the following steps of:
3.1 Resnet 8-based student network is built as a lightweight ship discriminator.
3.2 A knowledge distillation objective function of the student network is constructed and optimized by adopting a random gradient descent method:
wherein D L represents a set of marked samples, D U represents a set of unmarked samples, |D L | represents the number of marked samples, |D U | represents the number of unmarked samples, D represents one sample in the set, θ s is a trainable parameter of the student network, To monitor loss,/>For knowledge distillation regularization term, λ 2 represents the penalty coefficient.
Monitoring lossesThe method comprises the steps of category loss and regression loss, wherein the category loss is calculated by adopting a cross entropy function, and the regression loss is calculated by adopting a smoothl 1 function. The purpose of the knowledge distillation regular item is to migrate the knowledge learned by the teacher network to the student network, so that the student network can have the prediction of consistency with the teacher network, and the mathematical form is as follows:
Wherein f c represents a cross entropy function, f l represents a smooth-L1 function, f s represents a sample prediction function of a student network, f t represents a sample prediction function of a teacher network, f T represents a sharpening function, T represents a sharpening coefficient, and the mathematical form of the sharpening function is as follows:
where p represents a one-dimensional input vector, Represents/>, of the j-th component of pTo the power.
Fig. 3 is a graph of classification accuracy for teacher networks and Resnet-based student networks of different network structures. The figure illustrates the case where the three teacher network-student network combinations Resnet-Resnet 8, resnet-Resnet 8, and Resnet-Resnet 8 perform knowledge distillation in different training rounds. Wherein Resnet-Resnet 8 show that the teacher network and the student network are both based on the network structure of Resnet, which is used as a benchmark in experiments. It can be seen that Resnet110,110 and Resnet both act as teacher networks to improve the performance of their corresponding student networks. Since Resnet and Resnet are similar in performance, the corresponding student networks are similar in performance.
4) Converting the trained student network into ONNX format, adopting OpenVINO to further optimize and deploy the student network, and finally realizing rapid ship detection through the optimized student network, wherein the method comprises the following steps:
4.1 Format conversion and model deployment: the trained student network is converted into ONNX format, the precision of model parameters in ONNX format is converted into floating point number precision 16 by utilizing a OpenVINO tool kit, deployment and reasoning are carried out, and the speed of ship detection is further increased. The model size and calculation speed of the ship detection are shown in table 1.
Table 1 comparison of model size and calculation speed for ship detection
4.2 Area blocking: in remote monitoring, the position of the vessel in the image changes little in 1 second. In order to reduce the amount of computational redundancy, the input image is cut into 4 non-overlapping areas, and ship detection processing is performed on the 4 areas in a cyclic manner through the optimized student network, so that the speed of ship detection is further improved, and the detection result is shown in fig. 4. In the reasoning process, each 576×704 picture generates 2334 effective candidate frames through a sliding window, and the data in table 1 can be obtained, so that the time required for completing all areas is 2334/1000×74.7=174.3 ms, and only 174.3/4=43.6 ms is required for completing one area, thereby meeting the real-time ship detection performance.
The above examples are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above examples, and any other changes, modifications, substitutions, combinations, and simplifications that do not depart from the spirit and principle of the present invention should be made in the equivalent manner, and the embodiments are included in the protection scope of the present invention.
Claims (3)
1. A ship rapid detection method based on knowledge distillation and semi-supervised learning is characterized by comprising the following steps:
1) Constructing a ship data set by utilizing the existing ship data and the ship data collected by the deployed sea area, marking a ship target on part of the data in the data set, generating a training sample according to priori frame sliding and carrying out data enhancement, wherein the training sample comprises a marked sample and a non-marked sample;
2) Constructing a teacher network, pre-training by using a marked sample, and then performing semi-supervised training by combining a non-marked sample to realize ship detection;
constructing a teacher network and training, comprising the following steps:
2.1 Constructing a teacher network with large model capacity as a ship discriminator, and pre-training the network by using a labeling sample;
2.2 Semi-supervised objective function for constructing teacher network
Where D L represents the set of marked samples, D U represents the set of unmarked samples, |D L | represents the number of marked samples, |D U | represents the number of unmarked samples, D represents one sample in the set, [ theta ] t is a trainable parameter of the teacher network,To monitor loss,/>For a consistency regularization term, λ 1 represents a penalty coefficient;
The purpose of the consistency regular term is to enable different data enhancement samples of the same sample to have category consistency and improve the category aggregation degree of target features, and the mathematical form is as follows:
Where N represents the number of data enhancements, f c represents the cross entropy function, f t represents the sample prediction function of the teacher network, d (0) represents the original sample, d (i) represents the ith data enhanced sample of the original sample;
3) Constructing a lightweight student network, guiding the student network to realize knowledge distillation on marked samples and unmarked samples by using a trained teacher network, and comprising the following steps of:
3.1 A student network with less parameters is built as a lightweight ship discriminator;
3.2 Knowledge distillation objective function for constructing student network
Wherein D L represents a set of marked samples, D U represents a set of unmarked samples, |D L | represents the number of marked samples, |D U | represents the number of unmarked samples, D represents one sample in the set, θ s is a trainable parameter of the student network,To monitor loss,/>For knowledge distillation regularization term, λ 2 represents a penalty coefficient;
Monitoring losses The method comprises the steps of category loss and regression loss, wherein the category loss is calculated by adopting a cross entropy function, and the regression loss is calculated by adopting a smoothl1 function; the purpose of the knowledge distillation regular item is to migrate the knowledge learned by the teacher network to the student network, so that the student network can have the prediction of consistency with the teacher network, and the mathematical form is as follows:
Wherein f c represents a cross entropy function, f l represents a smooth-L1 function, f s represents a sample prediction function of a student network, f t represents a sample prediction function of a teacher network, f T represents a sharpening function, T represents a sharpening coefficient, and the mathematical form of the sharpening function is as follows:
where p represents a one-dimensional input vector, Represents/>, of the j-th component of pPower of the order;
4) And converting the trained student network into ONNX format, adopting OpenVINO to further optimize and deploy the student network, and finally realizing rapid ship detection through the optimized student network.
2. The method for rapid detection of a ship based on knowledge distillation and semi-supervised learning according to claim 1, wherein the step 1) comprises the steps of:
1.1 Acquiring existing ship data and acquiring data in a sea area to be deployed, constructing a ship data set, marking ship targets on part of the data to form marked data and unmarked data, and acquiring a multi-scale priori frame through clustering calculation of the ship targets in the marked data;
1.2 According to the sliding of the multi-scale prior frame, generating a multi-scale ship sample serving as a training sample, wherein the multi-scale ship sample comprises the following specific operations: sliding in the labeling data to generate a labeling sample, and dividing the labeling sample into a positive sample and a negative sample according to the intersection ratio of the sample and a labeling frame, wherein the positive sample comprises a multi-class ship target, records a class label and a position regression label of the multi-class ship target, and the negative sample is a background sample and slides in the non-labeling data to generate a non-labeling sample;
1.3 Data enhancement for training samples including horizontal flipping, color dithering, light variation, and random clipping.
3. The method for rapid detection of a vessel based on knowledge distillation and semi-supervised learning as set forth in claim 1, wherein the step 4) comprises the steps of:
4.1 Converting the trained student network into ONNX format, further optimizing the student network by OpenVINO and deploying;
4.2 The input image is cut into 4 non-overlapping areas, and the 4 areas are monitored circularly through the optimized student network, so that the ship detection speed is further improved.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111410408.2A CN114241285B (en) | 2021-11-25 | 2021-11-25 | Ship rapid detection method based on knowledge distillation and semi-supervised learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111410408.2A CN114241285B (en) | 2021-11-25 | 2021-11-25 | Ship rapid detection method based on knowledge distillation and semi-supervised learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114241285A CN114241285A (en) | 2022-03-25 |
| CN114241285B true CN114241285B (en) | 2024-05-28 |
Family
ID=80751110
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111410408.2A Active CN114241285B (en) | 2021-11-25 | 2021-11-25 | Ship rapid detection method based on knowledge distillation and semi-supervised learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114241285B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115239638A (en) * | 2022-06-28 | 2022-10-25 | 厦门微图软件科技有限公司 | Industrial defect detection method, device and equipment and readable storage medium |
| CN115082792B (en) * | 2022-06-29 | 2024-05-28 | 华南理工大学 | Cross-domain water surface target detection method based on feature countermeasure migration and semi-supervised learning |
| CN115082800B (en) * | 2022-07-21 | 2022-11-15 | 阿里巴巴达摩院(杭州)科技有限公司 | Image segmentation method |
| CN115019183B (en) * | 2022-07-28 | 2023-01-20 | 北京卫星信息工程研究所 | Remote sensing image model migration method based on knowledge distillation and image reconstruction |
| CN115018852B (en) * | 2022-08-10 | 2022-12-06 | 四川大学 | Abdominal lymph node detection method and device based on semi-supervised learning |
| CN115131747A (en) * | 2022-08-25 | 2022-09-30 | 合肥中科类脑智能技术有限公司 | Knowledge distillation-based power transmission channel engineering vehicle target detection method and system |
| CN116030023A (en) * | 2023-02-02 | 2023-04-28 | 泉州装备制造研究所 | Point cloud detection method and system |
| CN116168256B (en) * | 2023-04-19 | 2023-07-14 | 浙江华是科技股份有限公司 | Ship detection method, system and computer storage medium |
| CN117610421B (en) * | 2023-11-28 | 2024-09-24 | 四川大学 | Offshore wind power intelligent design application method for learning transfer after foresight distillation |
| CN118196145B (en) * | 2024-05-16 | 2024-07-09 | 云南师范大学 | Teacher classroom vision tracking method based on single-point labeling and weak semi-supervision |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112464879A (en) * | 2020-12-10 | 2021-03-09 | 山东易视智能科技有限公司 | Ocean target detection method and system based on self-supervision characterization learning |
| CN112529178A (en) * | 2020-12-09 | 2021-03-19 | 中国科学院国家空间科学中心 | Knowledge distillation method and system suitable for detection model without preselection frame |
| CN112766087A (en) * | 2021-01-04 | 2021-05-07 | 武汉大学 | Optical remote sensing image ship detection method based on knowledge distillation |
| CN113313166A (en) * | 2021-05-28 | 2021-08-27 | 华南理工大学 | Ship target automatic labeling method based on feature consistency learning |
-
2021
- 2021-11-25 CN CN202111410408.2A patent/CN114241285B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112529178A (en) * | 2020-12-09 | 2021-03-19 | 中国科学院国家空间科学中心 | Knowledge distillation method and system suitable for detection model without preselection frame |
| CN112464879A (en) * | 2020-12-10 | 2021-03-09 | 山东易视智能科技有限公司 | Ocean target detection method and system based on self-supervision characterization learning |
| CN112766087A (en) * | 2021-01-04 | 2021-05-07 | 武汉大学 | Optical remote sensing image ship detection method based on knowledge distillation |
| CN113313166A (en) * | 2021-05-28 | 2021-08-27 | 华南理工大学 | Ship target automatic labeling method based on feature consistency learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114241285A (en) | 2022-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114241285B (en) | Ship rapid detection method based on knowledge distillation and semi-supervised learning | |
| Chen et al. | Deep learning for autonomous ship-oriented small ship detection | |
| CN111639692B (en) | Shadow detection method based on attention mechanism | |
| CN111460968B (en) | Unmanned aerial vehicle identification and tracking method and device based on video | |
| CN110097075B (en) | Deep learning-based marine mesoscale vortex classification identification method | |
| Wang et al. | Tree leaves detection based on deep learning | |
| Han et al. | Research on multiple jellyfish classification and detection based on deep learning | |
| CN113255837A (en) | Improved CenterNet network-based target detection method in industrial environment | |
| Arya et al. | Object detection using deep learning: A review | |
| Intwala et al. | Indian sign language converter using convolutional neural networks | |
| CN113139896A (en) | Target detection system and method based on super-resolution reconstruction | |
| CN114187506B (en) | Remote sensing image scene classification method of viewpoint-aware dynamic routing capsule network | |
| CN114898290A (en) | Real-time detection method and system for marine ship | |
| Li et al. | Object detection for uav images based on improved yolov6 | |
| Shang et al. | Faster and lighter meteorological satellite image classification by a lightweight channel-dilation-concatenation net | |
| Li | Parallel two-class 3D-CNN classifiers for video classification | |
| Zhao et al. | Ocean ship detection and recognition algorithm based on aerial image | |
| Ma | Fixed-point tracking of English reading text based on mean shift and multi-feature fusion | |
| CN115578364A (en) | Weak target detection method and system based on mixed attention and harmonic factor | |
| CN116245157A (en) | Facial expression representation model training method, facial expression recognition method and facial expression recognition device | |
| Langenkämper et al. | COATL-a learning architecture for online real-time detection and classification assistance for environmental data | |
| Da et al. | Remote sensing image ship detection based on improved YOLOv3 | |
| Li | Investigation of you only look once networks for vision-based small object detection | |
| Karim et al. | Bangla Sign Language Recognition using YOLOv5 | |
| Li et al. | Underwater object detection based on improved SSD with convolutional block attention |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |