基于改进MobileNet V3网络的桃子成熟度分级方法

Peach Ripeness Grading Method Based on Improved MobileNet V3 Network

目前在我国桃业生产过程中主要采用基于主观经验的人工方式对桃子外观成熟度进行分级,该方式不仅效率较低,而且易受主观因素的影响,导致同一批次的桃子在成熟度等级上参差不齐,无法达到国际桃品销售中所要求的成熟度品级标准。针对上述问题,本研究提出一种基于改进卷积神经网络MobileNet V3的桃子外观成熟度分级模型CS-MobileNet-P-L:首先,为了提升模型的特征提取能力,将多方位协调注意力机制模块引入原有注意力机制中,以构成双重注意力机制;其次,为提高模型的分级准确度,对网络Bneck结构中的激活函数进行调整并对模型的Last Stage结构进行优化改进。结果表明,当使用相同训练策略及环境配置时,改进后的CS-MobileNet-P-L模型的准确度比MobileNet V3模型提高了2.71个百分点,能较好地实现桃子外观成熟度的自动化精准分级。

At present,in the peach production in China,the peach apparent ripeness is still mainly gra-ded by artificial method based on subjective experience.This method is not only less efficient,but also easily influenced by subjective factors,resulting in uneven maturity grades of the same batch of peaches,which is unable to meet the maturity grade standards required for international sales.To address the above problems,a peach apparent ripeness grading model CS-MobileNet-P-L was put forward in this paper based on the improved convolutional neural network MobileNet V3.Firstly,in order to improve the feature extraction ability of the model,a multidirectional coordinated attention mechanism module was introduced into the original attention mechanism to constitute a dual attention mechanism.Secondly,in order to improve the grading accuracy of the model,the activation function in the Bneck structure of the network was adjusted,and the Last Stage structure of the model was optimized and improved.The results showed that under the same training strategy and envi-ronment configuration,the accuracy of the improved CS-MobileNet-P-L model was 2.71 percentage points higher than that of the MobileNet V3 model,thus better realized the automated and accurate grading of peach apparent ripeness.