Traditional Full-Stokes detection methods are based on time-division or spatial-division approaches， which suffer from drawbacks such as large device size， challenging integration， and inability to detect consistently in space and time. Recent advancements in two-dimensional materials and metamaterials have made it possible to realize ultracompact， spatiotemporally coherent Full-Stokes detectors based on well-defined polarization-sensitive structures at subwavelength scales. This study proposes a polarization detector based on graphene-metal nanoantennas， which leverages vector photocurrents generated on graphene and a neural network algorithm for reconstruction， enabling spatiotemporally coherent Full-Stokes parameter detection. Vector photocurrents under different polarization states of incident light were obtained through FDTD simulations. Subsequently， a mapping relationship S^=f(I^) between the Full-Stokes parameters S^ and the recorded vector photocurrents I^ was established using a neural network algorithm， successfully enabling the detection of Full-Stokes parameters. At a wavelength of 4 μm， the mean square error was 0.007 69. The relative radius difference of the minimum enclosing spheres for the actual and predicted Stokes parameters of the incident light is 7.68%. This detector design offers a new approach for achieving more integrated and miniaturized spatiotemporally coherent Full-Stokes detection. This detector effectively overcomes the inherent technical bottlenecks of time-division and spatial-division mechanisms， offering a novel approach for achieving more integrated and miniaturized spatiotemporally consistent full-Stokes detection.