Infrared weak and small target detection method for decoupling target extraction and false alarm elimination
Through the infrared weak target detection method of decoupling target extraction and false alarm removal, the LCM algorithm and ResNet18 network structure are used to solve the problem of high false alarm rate in complex sea and sky environments, and effective detection of weak targets and effective removal of false alarms is achieved.
Patent Information
- Application Number
- CN202411935716.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-26
- Publication Date
- 2025-05-13
AI Technical Summary
In complex sea and sky environments, there are a large number of noises similar to the target in infrared weak target detection, resulting in a high false alarm rate and it is difficult to distinguish between target and background interference.
An infrared weak target detection method is adopted for decoupling target extraction and false alarm removal. The target detection is performed through the LCM algorithm, and 28×28 pixel images are cropped as the false alarm removal data set. The ResNet18 network structure is used for training to obtain a false alarm removal network that can distinguish between the target and the false alarm.
Effectively eliminate most false alarms, reduce false alarm rates, avoid problems that cannot be taken into account with detection rate and false alarm rates, and ensure effective detection of weak targets in complex environments.
Smart Images
Figure CN119992314A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of infrared small target detection, and in particular relates to an infrared small target detection method for decoupling target extraction and false alarm elimination. Background Art
[0002] Infrared small target detection aims to determine the position of the target in the low signal-to-noise ratio infrared image, which is the basis for applications such as sea rescue, military reconnaissance, and military early warning. Compared with radar detection, infrared imaging systems have the advantages of being completely passive, anti-interference, and having strong camouflage capabilities. Therefore, infrared target detection in complex sea and sky environments has become an important means of reconnaissance and early warning.
[0003] In military applications, targets are mostly non-cooperative surface ships, stealth aircraft, drones, tactical missiles, etc. Due to the long imaging distance, the imaging equipment is affected by factors such as atmospheric attenuation, light changes, and temperature changes. The target occupies few pixels in the image and has low contrast with the surrounding environment. It mostly appears as a weak target (Figure 1a), making it difficult to detect.
[0004] In addition, complex sea and sky environments such as waves, fish scales, sea-sky lines, and broken clouds lead to the presence of a large number of noise points similar to the target in the infrared reconnaissance and early warning images (Figure 1b), resulting in a high false alarm rate in the detection results, which seriously restricts the accuracy of downstream target tracking, trajectory establishment, intelligent recognition and other tasks. Therefore, the infrared target detection model is required to have low false alarm capability in complex sea and sky environments while ensuring the ability to detect weak targets. Summary of the invention
[0005] In view of the complex sea and sky environment, there are a large number of noise points with similar height to the target. The interferences such as broken clouds and waves present a pixel distribution similar to the target in the image, which makes it difficult to distinguish the target from the background interference only through the target detection network and leads to a high false alarm rate. The present invention provides an infrared weak small target detection method for decoupling target extraction and false alarm elimination.
[0006] The technical solution adopted by the present invention to solve the technical problem is: a method for detecting infrared weak small targets by decoupling target extraction and false alarm elimination, comprising the following steps: (1) Collect and annotate infrared small target image samples, construct an image dataset, and divide it into an infrared small target detection image training set and an image test set; (2) Use the LCM algorithm to perform target detection on the image training set and save the coordinates of the target center point as the detection result; (3) Based on the detection results, a 28×28 pixel image is cropped at the corresponding position of the image with the coordinates of the center point of each detection result as the training data of the false alarm rejection dataset; (4) According to the actual application scenario and based on the real labels, the cropping results are divided into target, false alarm i, and i+1 categories to construct a false alarm elimination dataset; (5) Based on the ResNet18 network structure, the false alarm rejection dataset was used for training to obtain a false alarm rejection network that can distinguish the above five types of targets in the LCM extraction results; (6) Use the LCM algorithm to perform target detection on the image test set. Based on the detection results, a 28×28 pixel image is cropped with the coordinates of the center point of each detection result as the center. This image is used as the false alarm rejection data set test data and input into the trained false alarm rejection network to obtain the target position.
[0007] Furthermore, in step (2), the mean and standard deviation thresholds of the LCM algorithm are set to 0.2 and 1.
[0008] Furthermore, in the step (4), the clipping results are divided into five categories: target, building false alarm, wave false alarm, sky false alarm, and tree false alarm.
[0009] The beneficial effect of the present invention is that compared with the existing method, the present invention can set a low threshold LCM algorithm to ensure the detection rate while effectively eliminating most false alarms, and effectively avoid the problem of the LCM algorithm being unable to balance the detection rate and false alarm rate due to the threshold setting problem. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] Figure 1 To prevent small targets and sea waves from interfering with the image; Figure 2 The present invention is a flow chart of the detection method. DETAILED DESCRIPTION
[0011] The present invention will be further described below in conjunction with specific embodiments and drawings.
[0012] The present invention proposes an infrared dim small target detection method in which a detection model and a false alarm rejection model are decoupled and paired, so as to alleviate the problem of high false alarm rate of the existing method in complex sea and sky scenes.
[0013] Reference Figure 2 As shown, the present invention discloses an infrared weak small target detection method for decoupling target extraction and false alarm elimination, which includes the following steps.
[0014] (1) Collect and annotate infrared small target detection images, construct an image dataset, and divide it into an infrared small target detection image training set and an image test set.
[0015] (2) Use the LCM algorithm to detect targets on the image training set and save the target center point coordinates as the detection results. In this step, the mean and standard deviation thresholds of the LCM algorithm are set to 0.2 and 1.
[0016] (3) Based on the detection results, a 28×28 pixel image is cropped at the corresponding position of the image with the coordinates of the center point of each detection result as the training data of the false alarm rejection dataset.
[0017] (4) According to the actual application scenario and based on the true labels, the cropping results are divided into target, false alarm i, and i+1 categories to construct a false alarm elimination dataset.
[0018] In this embodiment, the clipping results are divided into five categories: target, building false alarm, wave false alarm, sky false alarm, and tree false alarm.
[0019] (5) Based on the ResNet18 network structure, the false alarm rejection dataset was used for training to obtain a false alarm rejection network that can distinguish the above five types of targets in the LCM extraction results.
[0020] (6) Use the LCM algorithm to perform target detection on the image test set, crop the results according to step (3), input them into the trained false alarm rejection network, and obtain the target location.
[0021] The above embodiments are only illustrative of the principles and effects of the present invention, as well as some embodiments of its application. For those skilled in the art, several modifications and improvements may be made without departing from the creative concept of the present invention, and all of these belong to the protection scope of the present invention.
Claims
1. A method for detecting infrared small and weak targets by decoupling target extraction and false alarm elimination, characterized in that: The following steps are included (1) Collect and annotate infrared small target image samples, construct an image dataset, and divide it into an image training set and an image test set; (2) Use the LCM algorithm to perform target detection on the image training set and save the coordinates of the target center point as the detection result; (3) Based on the detection results, a 28×28 pixel image is cropped with the coordinates of the center point of each detection result as the training data of the false alarm rejection dataset; (4) Based on the true labels, the cropping results are divided into target, false alarm i, and i+1 categories to construct a false alarm elimination dataset; (5) Based on the ResNet18 network, the false alarm rejection dataset is used for training to obtain a false alarm rejection network that can distinguish the above targets; (6) Use the LCM algorithm to perform target detection on the image test set. Based on the detection results, a 28×28 pixel image is cropped with the coordinates of the center point of each detection result as the center. This image is used as the false alarm rejection data set test data and input into the trained false alarm rejection network to obtain the target position.
2. The infrared small target detection method for decoupling target extraction and false alarm elimination according to claim 1 is characterized in that: In step (2), the mean and standard deviation thresholds of the LCM algorithm are set to 0.2 and 1.
3. The infrared small target detection method for decoupling target extraction and false alarm elimination according to claim 2 is characterized in that: In the step (4), the clipping results are divided into five categories: target, building false alarm, wave false alarm, sky false alarm, and tree false alarm.