This paper proposes a method for the intelligent detection of high-frequency surface wave radar (HFSWR) targets. This method cascades the adaptive constant false alarm (CFAR) detector variability index (VI) with the convolutional neural network (CNN) to form a cascade detector (VI)CFAR-CNN. First, the (VI)CFAR algorithm is used for the first-level detection of the range–Doppler (RD) spectrum; based on this result, the two-dimensional window slice data are extracted using the window with the position of the target on the RD spectrum as the center, and input into the CNN model to carry out further target and clutter identification. When the detection rate of the detector reaches a certain level and cannot be further improved due to the convergence of the CNN model, this paper uses a dual-detection maps fusion method to compensate for the loss of detection performance. First, the optimized parameters are used to perform the weighted fusion of the dual-detection maps, and then, the connected components in the fused detection map are further processed to achieve an independent (VI)CFAR to compensate for the (VI)CFAR-CNN detection results. Due to the difficulty in obtaining HFSWR data that include comprehensive and accurate target truth values, this paper adopts a method of embedding targets into the measured background to construct the RD spectrum dataset for HFSWR. At the same time, the proposed method is compared with various other methods to demonstrate its superiority. Additionally, a small amount of automatic identification system (AIS) and radar correlation data are used to verify the effectiveness and feasibility of this method on completely measured HFSWR data.

This paper presents a novel method for clutter suppression and target detection on the Range-Doppler (RD) spectrum of High Frequency Surface Wave Radar (HFSWR). The approach replaces the traditional classification layer of the residual neural network with a regression layer, forming the Residual Regression Network (RRN). Subsequently, a target-embedded dataset is constructed using real measurement data and supplemented simulation data to train the network in learning the characteristics of targets, clutter, and noise interference in the RD spectrum of HFSWR, thus generating a spectrum containing only target information. Introducing an exponential weight control factor, the network output is exponentially weighted and Hadamard multiplied with the original input’s spectrum to obtain the final spectrum for detection. This final spectrum is then fed into the Constant False Alarm Rate (CFAR) detector for further analysis. Through the verification of multiple data sets of target embedded data and measured data and multiple experimental groups, this method demonstrates superior clutter suppression and target detection performance compared to existing methods. Operators can adjust the suppression level by modifying the exponential weight control factor and optimize detection performance by tuning the control factor and CFAR detector parameters according to specific operational requirements.

High-Frequency Surface Wave Radar (HFSWR) is a crucial tool for oceanic remote sensing and surveillance; however, radar target detection is challenged by the presence of background clutter and interference. In response, this paper designs a novel dual-scale branch fusion network specifically for detecting target signals in the range-Doppler (RD) spectrum. The network effectively enhances the ability to distinguish between targets and clutter by combining large-scale environmental feature sensing with small-scale target signal structure analysis. Additionally, we propose a novel detection threshold adjustment mechanism based on the RD spectrum perception network. First, an initial detection threshold is calculated using the traditional Constant False Alarm Rate (CFAR) method. Then, the output of the softmax layer in the RD spectrum perception network is used to adjust the threshold, improving the robustness and accuracy of the detection process. The RD spectrum perception network is trained jointly using data from the Automatic Identification System (AIS) associated with HFSWR and simulated target-embedded data. Multiple validations and analyses of the proposed detection method are conducted with these datasets. Experimental results demonstrate that the proposed method has good detection performance, outperforming several other existing methods.

Ship detection has been an area of focus for high-frequency surface wave radar (HFSWR). The detection and identification of ship formation have proven significant in early warning, while studies on the formation identification are limited due to the complex background and low resolution of HFSWR. In this paper, we first establish a spatial distribution model of ship formation in HFSWR. Then, we propose a cascade identification algorithm of ship formation in the clutter edge. The proposed algorithm includes a preprocessing stage and a two-stage formation identification stage. The Faster R-CNN is introduced in the preprocessing stage to locate the clutter regions. In the first stage, we propose an extremum detector based on connected regions to extract suspicious regions. The suspicious regions contain ship formations, single-ship targets, and false targets. In the second stage, we design a network connected by a convolutional neural network (CNN) and an extreme learning machine (ELM) to identify two densely distributed ship formations from inhomogeneous clutter and single-ship targets. The experimental results based on the factual HFSWR background demonstrate that the proposed cascade identification algorithm is superior to the extremum detector combined with the classical CNN algorithm for ship formation identification. Meanwhile, the proposed algorithm performs well in weak formation and deformed formation identification.

High-Frequency Surface Wave Radar (HFSWR) faces severe challenges from complex interference in maritime monitoring, hindering weak target detection and false alarm control. To address the model mismatch inherent in traditional Constant False Alarm Rate (CFAR) detectors and the limited perception and lack of sensitivity adjustability in existing encoder-decoder deep neural network methods, this paper proposes an RD Spectrum-Aware Network with Controllable Detection Sensitivity (RDSAN-CDS). The network adopts an end-to-end encoder-decoder architecture, integrating Transformer modules at the deep feature extraction stage to capture the global long-range dependencies between targets and the environment within the Range- Doppler (RD) spectrum. Furthermore, a gating mechanism driven by a preset false alarm probability is designed as an auxiliary input channel. This mechanism dynamically adjusts multi-scale feature responses during decoding, endowing the network with flexible detection sensitivity control during inference. Experimental results on field-measured radar data demonstrate that RDSAN-CDS outperforms existing mainstream methods such as Residual Regression Network (RRN) and DeepLabv3 in complex clutter backgrounds, achieving an effective balance and precise control of the detection rate and false alarm rate under diverse mission requirements.