Abstract: Coal-rock interface identification and automatic height adjustment are critical technologies for achieving intelligent and unmanned mining operations in longwall shearer systems. Traditional methods, such as manual observation or conventional techniques, are often employed for coal-rock interface identification. However, the shearers can't adapt to the complex changes of coal seam by adjusting the height through memory cutting and proportion-integration-differentiation (PID) control, and the accuracy of the control algorithm is limited. To address the challenges of intelligent height adjustment in shearers, this paper proposes an advanced shearer automatic height control scheme. A low-frequency impulse ultra-wideband (IR-UWB) electromagnetic wave radar is utilized, and the oblique projection operator based maximum likelihood like (OPMLL) method is employed for coal-rock interface identification. In this approach, the signal covariance matrix is respectively projected onto individual signal subspaces to generate several covariance matrices containing single-signal information. The maximum likelihood (ML) method is then applied to estimate the round-trip time delay. Moreover, the OPMLL method performs the iteration operation to further enhance estimation accuracy, ultimately yielding a precise trajectory of the coal-rock interface. Considering the specific requirements of coal mining process and equipment constraints, the cutting trajectory is optimized using a constrained optimization algorithm with the maximum recovery rate as the objective. Furthermore, based on the kinematics and state space model of shearer, a neural network observer based indirect adaptive performance control (NOIAPPC) method is developed for automatic height adjustment. Radial Basis Function Neural Networks (RBF NNs) are applied to estimate unknown nonlinear system functions, while the derivatives of the virtual control law at each step of backstepping control are estimated using a second-order filter. Additionally, system state variables are estimated online through a novel neural network sliding mode observer. As a result, by tuning the controller parameters, the proposed method ensures that both transient and steady-state errors remain within predefined bounds. Finally, the effectiveness and superiority of the proposed method are verified through experimental results, providing robust technical support for advancing the intelligence level of shearers. Experimental findings indicate that the angle tracking error of the shearer arm is maintained below 8.2%, the maximum error of the cutting trajectory does not exceed 0.01 m.