This paper considers measurement extraction for two closely-spaced objects with unknown equal intensities in an imaging sensor's focal plane array (FPA). Given a screen of FPA data, the first part of the measurement extractor, target location estimator, can extract the location estimates for two targets or one, with the corresponding accuracy given by the Cramér Rao lower bound (CRLB). The second part of the measurement extractor, target detector, selects among the hypotheses of two resolved targets and a single one using information-theoretic criteria and hypothesis tests. Simulation results have been conducted to evaluate the measurement extraction performance including the probability of resolving the two hypotheses, and the efficiency and unbiasedness of the target location estimates for the selected hypothesis using different hypothesis detection schemes. The generalized likelihood ratio test (GLRT) based on linearized observation model using second order Taylor series expansion is most appealing as it provides an explicit expression of the probability of detecting two targets as a function of the target separations, the signal-to-noise ratio at a given false resolution probability. It is shown that the simulation-based resolution performance for the GLRT using the estimated center location of the two targets matches well with the analytic performance assuming known center.

中文翻译：

---

使用成像传感器对两个相距很近的物体进行测量提取

本文考虑了对成像传感器焦平面阵列 (FPA) 中强度未知的两个相距很近的物体的测量提取。给定 FPA 数据的屏幕，测量提取器的第一部分，即目标位置估计器，可以提取两个或一个目标的位置估计值，相应的精度由 Cramér Rao 下限 (CRLB) 给出。测量提取器的第二部分，即目标检测器，使用信息论标准和假设检验在两个已解析目标和单个目标的假设中进行选择。已进行仿真结果以评估测量提取性能，包括解决两个假设的概率，以及使用不同假设检测方案对所选假设进行目标位置估计的效率和无偏性。基于使用二阶泰勒级数展开的线性化观察模型的广义似然比检验 (GLRT) 最吸引人，因为它提供了检测两个目标的概率作为目标分离、信噪比的函数的明确表达在给定的错误分辨率概率下。结果表明，使用两个目标的估计中心位置的 GLRT 基于模拟的分辨率性能与假设中心已知的分析性能非常匹配。基于使用二阶泰勒级数展开的线性化观察模型的广义似然比检验 (GLRT) 最吸引人，因为它提供了检测两个目标的概率作为目标分离、信噪比的函数的明确表达在给定的错误分辨率概率下。结果表明，使用两个目标的估计中心位置的 GLRT 基于模拟的分辨率性能与假设中心已知的分析性能非常匹配。基于使用二阶泰勒级数展开的线性化观察模型的广义似然比检验 (GLRT) 最吸引人，因为它提供了检测两个目标的概率作为目标分离、信噪比的函数的明确表达在给定的错误分辨率概率下。结果表明，使用两个目标的估计中心位置的 GLRT 基于模拟的分辨率性能与假设中心已知的解析性能非常匹配。