Real-time pallet detection method based on lightweight deep convolutional neural network

By using depth separation convolution and model pruning technology in pallet detection, the problem of excessive demand for deep convolution neural network computing resources in the existing technology is solved, real-time pallet detection on forklift industrial control machines is realized, and the needs of practical application scenarios are met.

CN119762868APending Publication Date: 2025-04-04YANGTZE RIVER DELTA HART ROBOT IND TECH RES INST
1 Cites 0 Cited by

Patent Information

Application Number
CN202411881474.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-12-19
Publication Date
2025-04-04

AI Technical Summary

Technical Problem

The existing deep convolutional neural networks have too high demand for computing resources, memory capacity and energy consumption in pallet detection, making it difficult to realize real-time detection on forklift industrial control machines with limited computing power.

Method used

By replacing the standard convolution with deep separable convolution and combining the network slimming algorithm and similarity evaluation method for model pruning, the computing resource requirements of the neural network are reduced and a lightweight model is obtained.

Benefits of technology

Real-time detection of pallets on forklift industrial control machines with limited computing resources is realized, meeting the needs of practical application scenarios, and improving the computing efficiency and accuracy of the model.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The present invention discloses a real-time pallet detection method based on a lightweight deep convolutional neural network, the method comprising selecting a neural network model, selecting the neural network model as a target detection model, modifying the standard convolution in the target detection model to a depth-separable convolution, training the modified target detection model, and then entering a pruning process after the training is completed to obtain a pruned target detection model, and making an accuracy judgment on the pruned target detection model. If the accuracy meets the requirements, the pruned model is deployed on a forklift to detect and identify the pallet; if the accuracy does not meet the requirements, the pruning rate is reduced and then fed back to the pruning process. By modifying and pruning the existing model, the demand for computing resources of the neural network model is reduced, and the modified lightweight model is used to identify the pallet, effectively meeting the actual application scenarios.
Need to check novelty before this filing date? Find Prior Art