Abstract: Diffusion magnetic resonance imaging (dMRI) is the only imaging method capable of noninvasively and quantitatively reconstructing the orientation of white matter fibers in the brain. Fiber tracking based on the fiber orientation distribution function (fODF) accurately represents the distribution of multiple fiber orientations, making it suitable for reconstructing complex fiber structures within brain tissue. However, fitting fODF typically relies on high-angular-resolution and multi-b-value dMRI data,which limits its clinical application to some extent. Deep learning, due to its powerful nonlinear mapping capabilities, has been employed to directly reconstruct high angular resolution fODF from conventional clinical low-b-value,low-angular-resolution dMRI signals. Yet,it faces challenges such as inability to handle diverse data inputs and a lack of domain-specific information. This study proposes a High Resolution Reconstruction Model (HRRM) by integrating deep learning networks with an Ordinary Differential Equation(ODE) solver. The model comprises two modules:1) a signal extraction module utilizing a 3D convolutional neural network to extract directional features; and 2) a high resolution reconstruction module that builds a neural network based on the Adams-Bashforth-Moulton (ABM) ODE solver to map low-angular-resolution fODF spherical harmonic coefficient features to their high-angular-resolution counterparts. The HRRM model directly generates high-angular-resolution fODF images from conventional clinical dMRI data, overcoming the limitations of low-resolution clinical data while enhancing the precision of white matter fiber tract analysis. This provides reliable technical support for research into various neurological disorders.