Abstract: Cognitive sensor local information sparse representation and compressive measurement are investigated, which are conducted by Analog-to-Information Converters (AIC) at each sensor in Cognitive Wireless Sensor Networks (C-WSN). Gradient Projection Sparse Reconstruction (GPSR) scheme based on energy-efficiency measurement is proposed. According to the spatial-temporal correlation structure of non-stationary signals perceived by massive cognitive sensors in Event Region (ER), these signals are mapped to wavelet orthogonal basis concatenate dictionaries to perform sparse representation. Adaptive measurement is implemented via weighted energy subset function, which could obtain the proper observation in energy-efficiency approach. The corresponding measurement matrix is constructed by the orthogonalization of these selected measurement vectors. Adaptive compressive reconstruction is performed at sink via GPSR algorithm, which is compared with conventional Orthogonal Matching Pursuit (OMP) algorithm. Simulation results indicate that, signal reconstruction effect based on energy-efficiency measurement GPSR adaptive compression is superior to Gaussian random measurement in the region where compression ratio is less than 0.2. With the same sensor numbers, the proposed GPSR adaptive compression approach has small reconstruction Mean Square Error (MSE) at low Signal-to-Noise Ratio (SNR) region, and the required measurement number is less than Gaussian random measurement, which guarantees sensors energy balance effectively.