基于卷积神经网络算法的苹果疤痕识别技术

Apple scar recognition technology based on convolutional neural network algorithm

针对在苹果外部品质检测中,传统方法存在识别精度低、速度慢等问题,本文提出了一种基于卷积神经网络算法的图像处理技术。首先,采取7种数据增强方法对原数据进行增强处理,用以扩展数据集,降低模型训练过拟合的风险;其次,对Faster R-CNN中的特征提取网络进行优化,采用K-means聚类算法优化RPN网络,使数据集重新聚类生成新的anchor长宽比设置,并增加模型对苹果疤痕信息的提取;最后,以原始Faster R-CNN模型作为对照模型,以识别精度和速度为评价指标进行试验。结果表明:基于EfficientNet B7特征提取的K-means-Eff-CNN模型准确度达到了94.7%,且能够在0.1 s内识别出苹果疤痕。

In view of the problems of low recognition accuracy and slow speed of traditional methods in apple’s external quality inspection,this paper proposes an image processing technology based on convolutional neural network algorithm.Firstly,seven data enhancement methods are used to enhance the original data to expand the data set and reduce the risk of overfitting in model training.Secondly,the feature extraction network in Faster R-CNN is optimized,and the RPN network is optimized by K-means clustering algorithm,so that the data set is re-clustered to generate a new anchor aspect ratio setting,and the model is added to extract apple scar information.Finally,the original Faster R-CNN model is used as the control model,and the recognition accuracy and speed are used as the evaluation indexes.The results show that the accuracy of k-means-Eff-CNN model based on EfficientNet B7 feature extraction reaches 94.7%,and apple scars could be identified within 0.1 s.