Abstract: ‍ ‍In scenarios involving wideband radar detection with high range resolution， background clutter exhibits increasingly strong non-Gaussian characteristics. As a typical non-Gaussian distribution model， the compound Gaussian model is reasonably used to describe the non-Gaussian statistical properties of clutter in practical environments. Unlike the point-like targets of traditional narrowband radars， wideband radar targets demonstrate extension among the radar radial range. Currently， many detectors designed for range-spread targets are mainly tailored to homogeneous Gaussian clutter environments， which makes them less applicable to practical detection scenarios. To address the issue of the insufficient utilization of information from clutter textures in the existing detectors of range-spread targets， especially under the background of compound Gaussian clutter with an unknown covariance matrix structure， this study models the texture component as an inverse Gamma distribution and investigates several adaptive subspace detectors for radar range-spread targets. According to the two-step design procedure and different test criteria， three adaptive subspace detectors are established. First， by assuming that the clutter texture component and covariance matrix structure are both known， three detection statistics are derived based on the Rao， Wald， and Gradient tests. Second， by solving the maximum a posteriori estimate of the unknown texture component and iterative estimate of the unknown covariance matrix structure and then substituting these estimates into the detection statistics obtained in the first step， three adaptive subspace detectors can be constructed for range-spread targets. Among them， the detection statistic of the proposed Rao test is interestingly consistent with that of the Gradient test. The theoretical results show that the three proposed detectors possess an asymptotic constant false alarm rate property with respect to the unknown covariance matrix structure， which serves as a pivotal criterion delineating the proficiency of the adaptive performance of a detector. Moreover， simulation experiments are conducted to analyze the effects of the clutter correlation， amount of training data， and models of multiple dominant scatterers on the detection performances of the designed detectors. The corresponding results indicate that both the proposed Rao-based and Wald-based detectors demonstrate enhanced detection performance as the amount of training data increases. Moreover， their detection performance is almost unaffected by the clutter one-lag correlation coefficient， which indicates the strong robustness of the detectors to clutter correlation variations. Furthermore， when the target energy gradually concentrates in only one range cell， the detection performance of the proposed Rao-based detectors decreases， suggesting that the Rao test is more sensitive to different models of multiple dominant scatterers. In contrast， the proposed Wald-based detector maintains relatively consistent detection performance across various models involving multiple dominant scatterers， showcasing superior robustness. Additionally， the numerical results show that the existing detectors designed for homogeneous Gaussian environments exhibit significant performance degradation in the compound Gaussian clutter， and the detection performance of the proposed detectors is also superior to those of the existing comparative detectors designed for compound Gaussian environments. Between two proposed detectors， the Wald-based detector slightly outperforms the Rao-based detector.