Convolutional Neural Network Intermediate Classifier for Processing Load Reduction

The patent divides the convolutional neural network into multiple stages with intermediate classifiers positioned between convolutional layers. Each intermediate classifier acts as a segmentation point that can independently evaluate and filter feature maps, allowing the system to process only relevant portions of the image data through subsequent layers, thereby reducing overall computational load while maintaining detection accuracy.

The intermediate classifiers perform preliminary evaluation of feature maps before they are processed by subsequent convolutional layers. By conducting this preliminary action at earlier stages, the system can identify and filter out non-relevant data early, preventing unnecessary computation in later layers and thus improving processing speed without sacrificing accuracy.

The patent segments the processing pipeline by introducing intermediate classifiers at multiple stages. These classifiers create decision points that can filter out non-relevant images early in the process, reducing the time spent processing unnecessary data through all convolutional layers while maintaining the ability to accurately detect pedestrians in relevant images.

The intermediate classifiers perform preliminary filtering actions before the main processing pipeline completes. By conducting this filtering at earlier stages, the system reduces the time required to process entire images through all layers, as many images can be discarded after the preliminary classification, thus reducing overall processing time without compromising detection accuracy.

The patent segments the feature map processing by introducing intermediate classifiers that can selectively evaluate and filter feature maps from different convolutional layers. This segmentation allows the system to consume computational resources more efficiently by processing only the feature maps that are most relevant for detection, rather than uniformly processing all feature maps through all layers.

The intermediate classifiers perform preliminary evaluation of feature maps to determine their relevance for pedestrian detection. By conducting this evaluation early, the system can filter out non-relevant feature maps before they are processed by subsequent layers, thereby reducing the total computational resources consumed while maintaining detection accuracy through the processing of only essential feature information.