长期以来，在存在干扰目标、杂波边缘等复杂环境中的雷达目标高性能检测问题一直是雷达信号处理技术领域的瓶颈性问题。如何在各种不同的复杂环境背景下，突破传统检测方法的诸多理想假设条件，对杂波实现高性能的智能化、精细化感知，以在有效控制虚警率的同时，提高复杂环境中目标的检测概率，是近年来一个重要的研究方向。恒虚警（Constant False Alarm Rate, CFAR）检测技术是雷达目标检测的常用技术，主要有均值类、有序统计量类、杂波图类和自适应类等CFAR技术。本文介绍了CA-CFAR、GO-CFAR、SO-CFAR、OS-CFAR和VI-CFAR五种现有CFAR检测器的原理和实现方式，仿真验证了这些算法的检测性能和虚警控制能力，并分析了每种检测器的适用场景和存在的问题。针对现有CFAR检测算法都只利用了雷达当前回波进行检测，而丢弃了大量历史数据的问题，本文先定义了杂波单元的特征概率密度函数（probability density function，PDF），然后提出了一种基于大量历史数据的改进k-means杂波分区算法。用特征PDF间的KL散度代替k-means算法中的欧氏距离，进行无监督杂波聚类。仿真和实测数据试验表明，这种无监督方法可以实现对多类杂波的准确分区，实现了对杂波的精细感知。针对复杂环境中的目标检测问题，在杂波精细感知的基础上，本文提出了一种基于KL散度的智能恒虚警目标检测算法KL-CFAR。通过参考单元间KL散度和杂波分区相结合，实现了选取尽可能最大量的独立同分布参考单元而去除非独立同分布的参考单元，性能逼近最优检测器。计算机仿真实验和实测数据测试都表明，KL-CFAR检测算法在多种检测场景下均能够对人工目标和实际目标都保持良好的检测性能，且具有较强的虚警控制能力，具有很好的实际应用潜力。

For a long time, the problem of high-performance detection of radar targets in complex environments such as jamming targets and clutter edges has always been a bottleneck in the field of radar signal processing technology. How to break through many ideal assumptions of traditional detection methods under various complex environmental backgrounds, and achieve high-performance intelligent and refined perception of clutter, so as to effectively control the false alarm rate and improve the detection probability of the target in complex environments is an important research direction in recent years.Constant False Alarm Rate (CFAR) detection technology is a common technology for radar target detection, mainly including mean value, ordered statistics, clutter map and adaptive CFAR technologies. This paper introduces the principles and implementations of five existing CFAR detectors, CA-CFAR, GO-CFAR, SO-CFAR, OS-CFAR and VI-CFAR. The detection performance and false alarm control ability of these algorithms are verified by simulation, and the applicable scenarios and existing problems of each detector are analyzed.Aiming at the problem that the existing CFAR detection algorithms only use the current radar echo for detection and discard a large amount of historical data, this paper first defines the characteristic probability density function (PDF) of the clutter unit, and then proposes an improved k-means clutter partitioning algorithm based on a large amount of historical data. The Kullback-Leibler divergence between feature PDFs is used to replace the Euclidean distance in the k-means algorithm for unsupervised clutter clustering. Simulation and measured data experiments show that this unsupervised method can achieve accurate partitioning of multiple types of clutter and achieve fine-grained perception of clutter.Aiming at the problem of target detection in complex environments, based on the fine perception of clutter, this paper proposes an intelligent constant false alarm target detection algorithm KL-CFAR based on KL divergence. Through the combination of KL divergence between reference units and clutter partitioning, it is possible to select the largest possible number of IID reference units and remove non-IID reference units, and the performance is close to the optimal detector. Both computer simulation experiments and measured data tests show that the KL-CFAR detection algorithm can maintain good detection performance for both artificial targets and actual targets in various detection scenarios, and has strong false alarm control capabilities and practical application potential.