Object recognition method and apparatus, and device and medium

An object recognition apparatus, wherein the apparatus is applied to a pre-constructed convolutional neural network for object recognition, comprising:

• a convolutional layer configured for extracting pre-selected feature information of objects to be recognized;
• a pooling layer configured for performing dimension reduction processing on the pre-selected feature information to obtain feature information of the objects to be recognized;
• and
• a fully connected layer configured for processing the feature information to obtain processed feature information, extracting a target feature value from the processed feature information, and compiling statistics on the number of objects to be recognized;
• wherein the fully connected layer is configured for determining, from pre-stored queue information, target queue information corresponding to the target feature value according to the target feature value, wherein the queue information comprises a queue ID, a queue length, a head address and a tail address of each queue;
• the head address is a storage address of reference feature information of a first sample object in the each queue;
• the tail address is a storage address of reference feature information of a last sample object in the each queue, and the each queue is established on a Double Data Rate Synchronous Dynamic Random Access Memory (DDR);
• and
• the fully connected layer is configured for performing comparative analysis on the processed feature information and reference feature information of each sample object in a queue corresponding to the target queue information and outputting a recognition result corresponding to the objects to be recognized;
• wherein that the fully connected layer is configured for determining, from pre-stored queue information, target queue information corresponding to the target feature value according to the target feature value, comprises:
• ranking the various pre-stored queue information according to the level of priority information in the queue information by the fully connected layer;
• and
• comparing the target feature value with a queue ID in each of the queue information after ranking by the fully connected layer in sequence, and taking the queue information with the same queue ID as the target feature value as target queue information corresponding to the target feature value.

A convolutional neural network, comprising:

• a convolutional layer configured for extracting pre-selected feature information of objects to be recognized;
• a pooling layer configured for performing dimension reduction processing on the pre-selected feature information to obtain feature information of the objects to be recognized;
• and
• a fully connected layer configured for processing the feature information to obtain processed feature information, extracting a target feature value from the processed feature information, and compiling statistics on the number of objects to be recognized;
• wherein the fully connected layer is configured for determining, from pre-stored queue information, target queue information corresponding to the target feature value according to the target feature value, wherein the queue information comprises a queue ID, a queue length, a head address and a tail address of each queue;
• the head address is a storage address of reference feature information of a first sample object in the each queue;
• the tail address is a storage address of reference feature information of a last sample object in the each queue, and the each queue is established on a Double Data Rate Synchronous Dynamic Random Access Memory (DDR);
• and
• the fully connected layer is configured for performing comparative analysis on the processed feature information and reference feature information of each sample object in a queue corresponding to the target queue information and outputting a recognition result corresponding to the objects to be recognized;
• wherein that the fully connected layer is configured for determining, from pre-stored queue information, target queue information corresponding to the target feature value according to the target feature value, comprises:
• ranking the various pre-stored queue information according to the level of priority information in the queue information by the fully connected layer;
• and
• comparing the target feature value with a queue ID in each of the queue information after ranking by the fully connected layer in sequence, and taking the queue information with the same queue ID as the target feature value as target queue information corresponding to the target feature value.