Aiming to address the challenge of identifying and eliminating systematic errors in various attitudes during target positioning of UAV optoelectronic reconnaissance platforms， a study was conducted on a target positioning system that integrates multi-dimensional attitude adaptive sensing algorithms. This paper presented a comprehensive target position calculation model that was based on the principles of target positioning. It utilized the optoelectronic reconnaissance platform's locking and tracking capabilities to perform multiple measurements of the cooperative target point. Furthermore， it analyzed the differentiated representations of the multi-dimensional attitude systematic error offset from the positioning results. Based on the principles of deep neural networks， the proposed model aimed to perform adaptive estimation and reverse compensation of systematic errors in a multi-task temporal feature extraction framework. Results of experiments indicate that when UAV operates at an altitude of 4 000 m， the system is able to mitigate 77% of systematic errors. It successfully reduces the target positioning error from 103 m to 19 m， resulting in an 81% increase in overall positioning accuracy. Consequently， this system enables high-precision positioning of UAV targets.