目标检测方法、目标检测模型的训练方法以及电子设备
By performing channel rearrangement and depth convolution processing on the color filter array image, target detection information is directly extracted, which solves the problem of insufficient detection efficiency and accuracy of color filter array images in edge devices, and realizes efficient and lightweight target detection.
CN116843910BActive Publication Date: 2026-07-17INST OF SEMICONDUCTORS - CHINESE ACAD OF SCI
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INST OF SEMICONDUCTORS - CHINESE ACAD OF SCI
- Filing Date
- 2023-05-19
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
In existing technologies, color filter array images are single-channel mosaic images, which require de-mosaic preprocessing, resulting in insufficient target detection efficiency and accuracy, and cannot be effectively applied, especially in edge devices with limited computing resources.
Method used
By rearranging the channels of the color filter array image, the first intermediate feature map is obtained. Then, through a series of convolution and residual convolution processes, the target detection information is directly generated, avoiding the need for mosaic preprocessing.
Benefits of technology
It improves the efficiency and accuracy of target detection in color filter array images and is suitable for edge devices with limited computing resources.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116843910B_ABST
Abstract
本公开提供了一种目标检测方法、目标检测模型的训练方法以及电子设备，可以应用于人工智能、计算机视觉、图像识别、图像传感器、模型压缩和目标检测技术领域。该目标检测方法包括：对彩色滤波阵列图像进行通道重排，得到第一中间特征图；根据第一中间特征图，得到第二中间特征图；根据第二中间特征图，得到至少一个尺度的第三中间特征图；以及，根据至少一个尺度的第三中间特征图，得到与至少一个待检测对象各自对应的目标检测信息。
Need to check novelty before this filing date? Find Prior Art