A method for detecting and classifying small targets at sea

Abstract

The present invention discloses a method for detecting and classifying small targets at sea, which belongs to the field of target recognition technology. In order to improve the detection and classification recognition of slow small targets in complex weather or environment, the effective separation of sea clutter background and slow small targets, effective modeling of slow small targets and improvement of signal-to-noise ratio, establishment of slow small target track and classification recognition and other technical research are carried out to improve the detection and classification recognition rate of slow small targets in rain, snow, haze, night and other conditions, assist in collision avoidance calculation of ships, and ensure navigation safety. Small fishing boats and small speedboats operating illegally are discovered in time, stable target information is provided to law enforcement personnel, and law enforcement efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of target recognition, and more specifically, relates to a method for detecting and classifying small targets at sea. Background Art

[0002] Small target intelligent detection and classification technology can intelligently detect the position and category of the target, and can quickly and accurately locate the target. It has strong application value in military fields such as drone reconnaissance, remote sensing images, and maritime target detection.

[0003] At present, target detection and classification technology has been widely used in civil fields such as monitoring systems. For example, on land, it detects and classifies people, animals, and vehicles in airports, stations, parking lots, residential areas and other places to maintain social order; at sea, it detects and identifies various large and small ships, buoys, floating objects and other targets to assist ships in navigation and collision avoidance calculations to ensure navigation safety.

[0004] However, in these fields, the diversity of targets and the mixing of targets and background environment in the acquired images or video information make it difficult to detect targets, and the existing detection and recognition algorithms in the civilian field are often difficult to apply. Therefore, it is of great research significance to reconstruct high-definition and high-contrast target information in complex environments, enhance the recognition technology of weak moving targets, and improve the target detection and recognition capabilities of intelligent systems in complex environments. Summary of the invention

[0005] In view of the above defects or improvement needs of the prior art, the present invention proposes a method for detecting and classifying small targets at sea. Aiming at the needs of detecting and classifying slow small targets on the sea surface in a complex sea clutter environment, the present invention studies new technologies such as effective separation of sea clutter background and slow small targets, effective modeling of slow small targets and improvement of signal-to-noise ratio, establishment of slow small target tracks and classification recognition, so as to improve the detection and classification performance of small targets on the sea surface.

[0006] To achieve the above object, the present invention provides a method for detecting and classifying small targets at sea, comprising:

[0007] Preprocess radar clutter data and optoelectronic video data, analyze the correlation and contrast between background data and images, and separate sea clutter background and slow small targets;

[0008] A deep learning algorithm is used to learn a large number of radar sea clutter background features measured in different environments, spaces and times, establish the amplitude and phase model of sea clutter, and extract the slow small target echo based on the amplitude and phase model;

[0009] Combined with the target optoelectronic video data, a prediction method based on deep learning is introduced to estimate the next position of the target, realize continuous tracking of the target, and ensure the continuity of the target track. After the target track is established, a deep learning method is used under the processing architecture of pre-tracking fusion to perform target characteristics, track classification and typical behavior analysis.

[0010] In some optional embodiments, the separation of the sea clutter background and the slow small target includes:

[0011] The multi-frame echo signals received during the dwell period of the radar scanning beam are accumulated. The newly measured radar echo data are combined with the previously accumulated radar echo data, and the VIBE improved method with a conservative update strategy and a deep learning method are used to obtain the separation structure and parameters of the sea clutter background and slow small targets, and to extract the clutter background features.

[0012] In some optional embodiments, the step of obtaining the separation structure and parameters of the sea clutter background and the slow small target includes:

[0013] The background model is established by using the improved VIBE method with a conservative update strategy. A sample set is stored for each pixel point of the accumulated radar echo data. Then, the new pixel value of each newly measured radar echo data is compared with the sample set to determine whether it belongs to the background point. If the new pixel value meets the preset requirements with the sampling value in the sample set, the new pixel value is considered to be a background point. The new pixel value is added to the background model to obtain the separation structure and parameters of the sea clutter background and slow small targets. Then, a large number of background model samples are learned using a deep learning method to extract the sea clutter background features and provide parameters for the modeling of slow small targets.

[0014] In some optional implementation schemes, the establishing of the amplitude and phase model of sea clutter and the extraction of slow small target echoes according to the amplitude and phase model include:

[0015] The method of pre-tracking fusion and accumulation averaging is adopted, and deep learning is used to establish the amplitude and phase model of sea clutter. According to the cumulative processing, the target signal is amplitude addition, and the noise is power addition. The amplitude addition accumulates faster than the power addition. After averaging, the variance of the noise decreases, the amplitude value of the target exceeds the threshold, and the slow small target echo is detected.

[0016] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0017] In order to improve the detection and classification of slow small targets in complex weather or environment, we have carried out effective separation of sea clutter background and slow small targets, effective modeling of slow small targets and improvement of signal-to-noise ratio, slow small target track establishment and classification and recognition technology research, improve the detection and classification recognition rate of slow small targets in rain, snow, haze, night and other conditions, assist in ship navigation collision avoidance calculation, and ensure navigation safety. Timely discover small fishing boats and small speedboats that are operating illegally, provide law enforcement personnel with stable target information, and improve law enforcement efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a system composition diagram of a method for detecting and classifying small targets at sea provided by an embodiment of the present invention;

[0019] Figure 2 This is a working principle diagram of a method for detecting and classifying small targets at sea provided by an embodiment of the present invention;

[0020] Figure 3 This is a workflow diagram of a method for detecting and classifying small targets at sea provided by an embodiment of the present invention;

[0021] Figure 4 It is a software interface diagram provided by an embodiment of the present invention;

[0022] Figure 5 This is a simulation effect diagram provided by an embodiment of the present invention, wherein (a) indicates radar X height 10m; sea condition 3; rain 0mm / h; RCS 7.5, 10, 5, 1, 0.1m 2 Simulation calculation, (b) indicates radar X height 10m; sea state 3; rain 4mm / h; RCS 7.5, 10, 5, 1, 0.1m 2 Simulation calculation, (c) indicates radar X height 10m; sea state 3; rain 16mm / h; RCS 7.5, 10, 5, 1, 0.1m 2 Simulation calculation. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0024] The present application provides a method for detecting and classifying small targets at sea. In order to improve the detection and classification of slow small targets in complex weather or environment, the present application conducts technical research on effective separation of sea clutter background and slow small targets, effective modeling of slow small targets and improvement of signal-to-noise ratio, establishment of slow small target tracks and classification and recognition, etc., specifically including:

[0025] (1) Effective separation of sea clutter background and slow small targets

[0026] In actual application scenarios, due to unpredictable changes in the external environment, we may encounter severe weather such as high sea conditions, strong storms, rain, snow, haze, etc., or insufficient light at night, drastic changes in light, etc., and there are often large differences between the radar clutter data and the target optoelectronic video data collected before and after. Therefore, before detecting small targets, it is necessary to preprocess the radar clutter data and optoelectronic video data to analyze the correlation and contrast between the background data and the image. The background model is established using the improved VIBE method (using a conservative update strategy), and a sample set is stored for each pixel point. Then, each new pixel value is compared with the sample set to determine whether it belongs to the background point. If the new observation value is close to the sample value in the sample set, it is considered to be a background point and added to the background model. The separation structure and parameters of the sea clutter background and slow small targets are obtained, and then a deep learning algorithm is used to learn a large number of background model samples, extract the sea clutter background features, and provide parameters for slow small target modeling.

[0027] (2) Effective modeling of slow and small targets and improving signal-to-noise ratio

[0028] The radar detection of slow large targets is different from that of slow small targets. The echo intensity of the former is much greater than that of sea clutter, and the signal-to-noise ratio is high, so it is easy to detect. The echo intensity of the latter is equivalent to or even weaker than that of sea clutter, and the signal-to-noise ratio is low, so it is not easy to detect. Therefore, the fundamental problem is how to improve the signal-to-noise ratio at the target position.

[0029] A deep learning algorithm is used to learn the background features of a large number of radar sea clutter measured in different environments, spaces, and times, and to establish an amplitude and phase model of sea clutter. Radar echoes that are different from this model may be slow small target echoes. Since the target signal in the cumulative processing is the addition of amplitudes, and the noise (irrelevant before and after) is the addition of powers, the amplitude addition accumulates faster than the power addition, and the signal can easily exceed the threshold. After averaging, the variance of the noise is reduced, and the amplitude value of the target exceeds the threshold, and the target can be detected. Therefore, the signal-to-noise ratio of the target can be improved by cumulative averaging.

[0030] (3) Track establishment and classification of slow small targets

[0031] Since the detection sequence of slow small targets may be intermittent and discrete, it is necessary to formulate new track establishment rules according to their characteristics. Here, combined with the target optoelectronic image information, a prediction algorithm based on deep learning is introduced to estimate the next position of the target and feed it back to the detection part to achieve continuous tracking of the target and ensure the continuity of the target track.

[0032] After the target track is established, a deep learning algorithm is used under the pre-tracking fusion processing architecture to analyze the target's characteristics, track classification, and typical behavior.

[0033] like Figure 1 As shown in the figure, the hardware composition diagram involved in the intelligent detection and classification method of small targets at sea includes optoelectronic evidence collection equipment, navigation radar, AIS, GPS / Beidou and integrated display and control console (including signal acquisition module and information processing module). Through the integrated design of radar, optoelectronic evidence collection equipment, Beidou and AIS, the target information is centrally displayed to form an organic whole, providing strong information support for ship visualization navigation and marine law enforcement surveillance and evidence collection.

[0034] like Figure 2 The figure shows its working principle diagram. The radar conducts all-round search and detection of the observed area, and sends the radar original echo signal to the radar signal acquisition module through the radar video input module. Based on modern radar detection technology, after radar video signal processing and multi-frame joint processing, a mathematical model for extracting and tracking slow small targets under strong sea clutter background is established, focusing on solving the detection and tracking of slow small targets under complex sea conditions and improving target extraction capabilities. Figure 3 The specific extraction process is as follows:

[0035] (1) Accumulate multiple frames of echo signals (including clutter and target) received during the radar scanning beam dwell period;

[0036] (2) The newly measured radar echo data is combined with the previously accumulated data, and the improved VIBE method and deep learning algorithm are used to obtain the separation structure and parameters of the sea clutter background and slow small targets, extract the clutter background features, and realize the intelligent detection of slow small targets with subsequent target track feedback;

[0037] (3) In the process of intelligent detection, some slow and small targets still cannot be detected normally. It is necessary to use optoelectronic and infrared information, adopt the method of pre-tracking fusion and accumulation averaging, and use deep learning to establish a multi-dimensional dynamic model of sea clutter to further improve the target signal-to-noise ratio in the detection process, so as to increase the detection probability of the target;

[0038] (4) In the measurement sequence, the slow small target sequence is detected intermittently, and it is necessary to improve the track establishment criteria, combine the optoelectronic image target information, add the prediction process, realize the track establishment of the slow small target, ensure the continuity of the target track, and predict the next step of the target track. Its information can be fed back to step (2);

[0039] (5) Based on the established target track, a deep learning algorithm is used under the processing architecture of pre-tracking fusion to analyze the target’s characteristics, track classification, and typical behavior, to further improve the recognition and classification of slow and small targets.

[0040] Furthermore, the system automatically performs information fusion processing on multi-source target data such as radar and AIS, establishes a comprehensive target track, tracks and captures various types of targets in the area, and provides a comprehensive target situation at sea. In addition, it continuously provides the location information of the target of interest to the optoelectronic evidence collection equipment (combined with the location information given by GPS / Beidou, the absolute position is inferred), displays the real-time video image information obtained by the optoelectronic evidence collection equipment, guides the optical equipment to perform optical detection on key directions and key targets, grasps the action details of the target, and cooperates with the radar to identify the target.

[0041] like Figure 4 The following is a typical simulation diagram of the interface of the slow small target tracking software. The software uses the electronic chart as the base map, and superimposes the navigation radar echo map to display the echo of the shore and sea targets. In addition, the navigation track of the slow small target is displayed in real time, and the video information of the photoelectric evidence equipment on the slow small target can be displayed in a window.

[0042] like Figure 5 As shown, (a) indicates radar X height 10m; sea state 3; rain 0mm / h; RCS 7.5, 10, 5, 1, 0.1m 2 Simulation calculation, (b) indicates radar X height 10m; sea state 3; rain 4mm / h; RCS 7.5, 10, 5, 1, 0.1m 2 Simulation calculation, (c) indicates radar X height 10m; sea state 3; rain 16mm / h; RCS 7.5, 10, 5, 1, 0.1m 2 Simulation calculation. Under the conditions of radar X height of 10m, sea state level 3, and rainfall of 0mm / h, 4mm / h, and 16mm / h, the typical target RCS is 7.5, 10, 5, 1, and 0.1m. 2 The results of simulation detection calculation. It can be seen from the figure that under various rainfall conditions, the RCS7.5, 10, 5m 2 All targets can be stably extracted and identified, with a maximum identification distance of more than 7 kilometers. In addition, RCS1, 0.1m 2 The target is greatly affected by rainfall and cannot be stably extracted and identified when the distance is greater than 3 kilometers.

[0043] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0044] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting and classifying small targets at sea, characterized in that: include: Preprocess radar clutter data and optoelectronic video data, analyze the correlation and contrast between background data and images, and separate sea clutter background and slow small targets; A deep learning algorithm is used to learn a large number of radar sea clutter background features measured in different environments, spaces and times, establish the amplitude and phase model of sea clutter, and extract the slow small target echo based on the amplitude and phase model; Combined with the target optoelectronic video data, a prediction method based on deep learning is introduced to estimate the next position of the target, realize continuous tracking of the target, and ensure the continuity of the target track. After the target track is established, a deep learning method is used under the processing architecture of pre-tracking fusion to analyze the characteristics, track classification and typical behavior of the target; The step of establishing the amplitude and phase model of sea clutter and extracting the slow small target echo according to the amplitude and phase model comprises: The method of pre-tracking fusion and accumulation averaging is adopted, and deep learning is used to establish the amplitude and phase model of sea clutter. According to the cumulative processing, the target signal is amplitude addition, and the noise is power addition. The amplitude addition accumulates faster than the power addition. After averaging, the variance of the noise decreases, the amplitude value of the target exceeds the threshold, and the slow small target echo is detected.

2. The method according to claim 1, characterized in that The separation of the sea clutter background and the slow small target includes: The multi-frame echo signals received during the dwell period of the radar scanning beam are accumulated. The newly measured radar echo data are combined with the previously accumulated radar echo data, and the VIBE improved method with a conservative update strategy and a deep learning method are used to obtain the separation structure and parameters of the sea clutter background and slow small targets, and to extract the clutter background features.

3. The method according to claim 2, characterized in that The method of obtaining the separation structure and parameters of the sea clutter background and the slow small target includes: The background model is established by using the improved VIBE method with a conservative update strategy. A sample set is stored for each pixel point of the accumulated radar echo data. Then, the new pixel value of each newly measured radar echo data is compared with the sample set to determine whether it belongs to the background point. If the new pixel value meets the preset requirements with the sampling value in the sample set, the new pixel value is considered to be a background point. The new pixel value is added to the background model to obtain the separation structure and parameters of the sea clutter background and slow small targets. Then, a large number of background model samples are learned using a deep learning method to extract the sea clutter background features and provide parameters for the modeling of slow small targets.

Images