A Small Target Detection Method Based on Two-Parameter KNN in Marine Environment

Abstract

This invention discloses a small target detection method based on a two-parameter KNN in a marine environment. Addressing the issue of the traditional KNN algorithm's inability to control the false alarm rate, this invention achieves this goal by adjusting the k value and the scaling parameter η. It represents an extension of pattern recognition technology or theory into the field of slow-moving small target detection in marine radar. It can delve deeper into the small target detection problem, exhibits excellent generalization capabilities, and is a novel and promising method for small target detection.

Description

Technical Field

[0001] This invention belongs to the field of slow-moving small target detection technology of maritime radar. The method uses a two-parameter KNN to detect small targets at sea based on the combination of multi-dimensional features. Background Technology

[0002] Small targets on the sea surface, primarily referring to ice floes, small boats, frogmen, and aircraft wreckage, are key and challenging targets for early warning, detection, and tracking by maritime radar. Slow-moving small targets emit weak echoes. Small sea surface targets have very small radar cross sections (RCS) and weak radar echoes, making them difficult to detect using traditional methods under low SCR conditions. Previous adaptive detection algorithms were mainly used for wide-area early warning and surveillance radars. However, sea clutter has a wide Doppler bandwidth; when sea surface targets are slow or small, their echoes are often submerged in strong sea clutter. Therefore, the performance of adaptive detection algorithms degrades significantly for weak sea surface targets. To address this problem, high-range resolution and long-term accumulated feature detection algorithms for weak sea surface targets have emerged. Researchers extract features in the time domain, frequency domain, and time-frequency domain, projecting samples into a more discriminative feature space, and then using pattern recognition and other methods for classification.

[0003] In ordinary binary classification problems, the probabilities of misclassification for the two classes are equivalent. However, in object detection, the probability of a false alarm is more important than the probability of a false miss (typically, the false alarm probability should be less than 10). -3 However, the false alarm rate can be as high as a few tenths, and we want it to be controllable. Therefore, small target detection algorithms based on two-parameter KNN have become another effective way to solve this problem. Existing algorithms include convex hull algorithms based on three features and controllable false alarm SVM, etc. Summary of the Invention

[0004] This invention addresses the shortcomings of existing technologies. Convex hull algorithms suffer from feature limitations, becoming difficult to compute when there are more than three features. SVM requires selecting a suitable kernel function and is computationally complex. This invention proposes a method based on a two-parameter KNN. The core technology of this invention is controllable false alarm rate. Using this invention, the false alarm rate can be controlled, improving detection accuracy. This invention represents an innovative development in the application of marine radar for detecting slow, small targets.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] Step 1: Feature extraction; extract time-domain and frequency-domain features as needed to construct a feature vector;

[0007] Step 2: Calculate the Euclidean distance between each test feature vector and all training samples to obtain the distance vector D;

[0008] Step 3: Arrange the elements in the distance vector D in ascending order, and take the k training samples corresponding to the k smallest distance values ​​to form the k nearest neighbors;

[0009] Step 4: Calculate the number of sea clutter samples and the number of target samples p and q in the k nearest neighbors, where p + q = k;

[0010] Step 5: Find the optimal k through grid optimization. * and η * ;

[0011] Step Six: For each test sample, in k * Among the nearest neighbors, if η * If p ≥ q, the test sample is determined to be the marine clutter label H0; otherwise, it is determined to be the target label H1.

[0012] Preferably, the feature extraction involves: extracting the required features from the training sample set and the test sample set to form a feature vector, and denoting the sea clutter label in the training sample as H0 and the target label as H1.

[0013] As a preferred method, the optimal k is found through grid optimization. * and η * Specifically:

[0014] Set the range of grid optimization to k∈[k1,k2], η∈[η1,η2], and optimize to obtain the optimal parameter k. * ,η * This makes the false alarm probability P realized fa * And the preset false alarm probability P fa The absolute value of the difference between them is less than a minimum value, and the resulting detection probability P d maximum.

[0015] The advantages of this invention compared to existing technologies are as follows: This invention addresses the problem of the traditional KNN algorithm's inability to control the false alarm rate. By adjusting the k value and the scaling parameter η, it achieves the goal of controlling the false alarm rate, representing an extension of pattern recognition technology or theory into the field of slow-moving small target detection in maritime radar. It can delve deeper into the small target detection problem, exhibits excellent generalization capabilities, and is a new and promising method for small target detection. Attached Figure Description

[0016] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0017] The following is in conjunction with the appendix Figure 1The following description illustrates specific embodiments of the present invention. It is for illustrative and explanatory purposes only and does not impose any limitations on the present invention.

[0018] A method for small target detection in sea clutter based on controllable false alarm KNN.

[0019] Step 1: Extract the required features from both the training and test sample sets to form feature vectors. Denote the sea clutter label in the training samples as H0 and the target label as H1.

[0020] Step 2: Calculate the Euclidean distance between each test feature vector and all training samples to obtain the distance vector D.

[0021] Step 3: Arrange the elements in the distance vector D in ascending order, and take the k training samples corresponding to the k smallest distance values ​​to form the k nearest neighbors.

[0022] Step 4: Calculate the number of sea clutter samples and the number of target samples p and q in the k nearest neighbors, where p + q = k.

[0023] Step 5: Set the proportional parameter η. If η·p≥q, then determine it as H0; otherwise, determine it as H1.

[0024] Step 6: Set the range of grid optimization k∈[k1,k2], η∈[η1,η2], and optimize to obtain the optimal parameter k. * ,η * This makes the false alarm probability P realized fa * And the preset false alarm probability P fa The absolute value of the difference between them is less than a minimum value, and the resulting detection probability P d maximum.

[0025] Step 7: For each test sample, in k * Among the nearest neighbors, if η * If p ≥ q, the test sample is classified as H0; otherwise, it is classified as H1.

[0026] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications or equivalent changes made to the above examples based on the algorithmic essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A small target detection method based on two-parameter KNN in a marine environment, characterized in that, The method specifically includes the following steps: Step 1: Feature extraction; extract time-domain and frequency-domain features as needed to construct a feature vector; Step 2: Calculate the Euclidean distance between each test feature vector and all training samples to obtain the distance vector D; Step 3: Arrange the elements in the distance vector D in ascending order, and take the k training samples corresponding to the k smallest distance values ​​to form the k nearest neighbors; Step 4: Calculate the number of sea clutter samples and the number of target samples p and q in the k nearest neighbors, where p + q = k; Step 5: Find the optimal k through grid optimization. * and η * Specifically: Set the range of grid optimization to k∈[k1,k2], η∈[η1,η2], and optimize to obtain the optimal parameter k. * ,η * This makes the false alarm probability P realized fa * And the preset false alarm probability P fa The absolute value of the difference between them is less than a minimum value, and the resulting detection probability P d maximum; Step Six: For each test sample, in k * Among the nearest neighbors, if η * If p ≥ q, the test sample is determined to be the marine clutter label H0; otherwise, it is determined to be the target label H1.

2. The method for small target detection in a marine environment based on a two-parameter KNN as described in claim 1, characterized in that: The feature extraction described herein is as follows: extracting the required features from the training sample set and the test sample set to form a feature vector, and denoting the sea clutter label in the training sample as H0 and the target label as H1.

Images