针对均匀数据场GPU光线投射技术中梯度、代理面和预积分计算存在的一些问题和多块绘制中可能存在缝隙的问题，设计了改进的GPU光线投射方法。主要改进有按需在物理空间实时计算梯度，估计梯度上界；单趟GPU预积分；设计体裁剪方法、缝隙消除方法。实验结果表明，改进的GPU光线投射技术便于采用梯度量调制，以突出感兴趣的物质; 可以交互修改转换函数，快速发现感兴趣的特征; 可以避免绘制错误；还可以逐步剥离前面的物体，看到后面的物体；可以对数据进行漫游; 绘制多块数据时没有缝隙。

Some problems exist in computing gradient, proxy polygon and pre-integration in GPU ray-casting for uniform grid and cracks may be easily produced in multi-block GPU ray-casting. To address these issues, improved GPU ray-casting for uniform grid is designed and implemented in this paper. Main improvements include an improved gradient computing method where that gradient is computed on need and on physical space and upper limit of gradient magnitude is evaluated, single pass GPU pre-integration, volume clipping method that is utilized to compute proxy polygon correctly and crack-free method for multi-block GPU ray-casting. Improved GPU ray-casting in this paper facilitates gradient magnitude modulation so that the interesting substances can be highlighted. It can compute pre-integration rapidly so that one can modify transfer function interactively and interesting features can be found quickly. Volume can be rendered correctly with it and one can peel the front thing and see the internal things. Crack is free when it renders multi-block dataset.