High-Efficiency Video Coding (HEVC/H.265) is the current standard for video coding. It reduces the bit rate by 50% in comparison to H.264. The process of intra-frame and inter-frame prediction in HEVC is significantly more complex and time-consuming. The Coding Tree Unit (CTU) Partition is the most significant component of HEVC intra and interframe prediction. Based on visual information, frames are divided into CTUs, with sizes ranging from 8 \(\times \) 8 to 64 \(\times \) 64 pixels. The partitioning of CTU in HEVC is determined by a recursive rate distortion optimisation procedure. Conversely, machine learning algorithms that are implemented in literature also use an iterative approach to partition the coding tree into distinct segments. The complexity of HEVC is significantly increased by this procedure, which is challenging. This paper presents a high-speed CU LeNet based deep learning architecture for resolving the recursive search problem. The proposed high-speed CU LeNet architecture accelerates CTU partitioning in intra-frame configurations by eliminating the need for iterative procedures, outperforming existing HEVC methods. In comparison to conventional HEVC, simulation results indicate that the modified LeNet-5 reduces the time required to encode a variety of test videos by approximately 87.80%.
