基于MobileNetV3-MHSA的输送带接扣损伤检测

Conveyor Belt Buckle Damage Detection Based on MobileNetV3-MHSA

针对带式输送机在运行时出现带扣断裂的问题,提出一种基于Mobile NetV3-MHSA的带扣损伤检测模型。通过限制对比度自适应直方图均衡化(CLAHE)算法进行图像增强,提高弱光环境下的对比度。对MobileNetV3进行改进,引入多头自注意力(MHSA)机制构造新的Bneck模块,在网络输出的全连接层中添加Softmax激活函数,采用Radam在模型训练初期对整体优化过程中的梯度信息进行调整,从而构造带扣损伤检测模型,进一步提高分类准确率。实验结果表明,改进后的算法分类准确率达到90.45%,相比于MobileNetV3提高4.15%,能够胜任带扣损伤检测的任务。

In order to solve the problem of belt buckle breakage during operation of belt conveyor,belt buckle damage detection model based on MobileNetV3-MHSA was proposed.The contrast limited adaptive histogram equalization(CLAHE)algorithm was used for image enhancement to improve the contrast in low light environment.Improved the MobileNetV3,and the multi-head self attention(MHSA)mechanism was introduced to construct a new Bneck module.Softmax activation function was added to the fully connected layer of network output,and gradient information in the overall optimization process was adjusted by Radam at the initial stage of model training.Therefore,a model for damage detection of belt buckle was constructed to further improve the classification accuracy.The experimental results show that the classification accuracy of the improved algorithm reaches 90.45%,which is 4.15%higher than that of MobileNetV3,and is competent for the task of detecting the damage of belt buckle.