Convolutional Neural Network Edge-Server Split for Image Recognition

The patent divides the convolutional neural network into multiple layers and segments the processing between mobile device and server. Specifically, lower layers (first to third layers) are processed on the mobile device to extract initial features, while upper layers (fourth to sixth layers) are processed on the server to perform final recognition. This segmentation reduces the computational burden on the mobile device while maintaining recognition accuracy through coordinated processing across both systems.

The patent applies preliminary action by having the mobile device process the first three layers of the convolutional neural network before transmitting data to the server. This preliminary processing extracts essential features from the image data, reducing the amount of data that needs to be transmitted to the server while ensuring that the server receives pre-processed information that maintains recognition accuracy. This approach minimizes transmission time and network traffic.

The patent replaces the conventional mechanical feature point descriptor search approach with a convolutional neural network-based system. Instead of using traditional computer vision algorithms that rely on hand-crafted features and geometric transformations, the system uses learned features from convolutional layers that are inherently more robust to transformations like rotation. The neural network automatically learns rotation-invariant features through its hierarchical structure, eliminating the need for complex geometric processing and improving reliability under various image transformations.