A sandwhich-typed flexible sensor has been designed based on the combination of the anisotropic properties of the double network polyaniline/polyvinyl alcohol composite hydrogel (PANI/PVA), which is mainly used for a composite measurement of complex stress in pipelines. Firstly, a finite element software is used to verify the measurability of the sensor for three-dimensional spatial forces, with a calibration platform established as well. Due to such characteristics of flexible sensors as individualized parameter difference and force dimension coupling, 500 groups of data are collected on the calibration platform, covering repeated experiments in multiple directions and scales. A BP neural network algorithm optimized by sparrow search algorithm has been designed to decouple the test results, with four voltage signals of the sensor successfully decoupled into the force condition of the sensor vector. The results show that compared with a single BP neural network algorithm, the proposed algorithm is characterized with an improved stability and a faster convergence speed, not easy to fall into local extremum with an improved performance in the detection of three-dimensional force.