基于改进的卷积神经网络水果分类算法设计

Design of fruit classification algorithm based on improved convolutional neural network

水果分类对于水果生产、加工、运输以及自助销售都有重要意义.卷积神经网络通过多层次的特征学习和自动特征提取,能够高效地处理和分类大量水果,在水果分类方面具有出突出的优势.然而,目前的水果分类方法存在诸多问题,如过度依赖人工、准确率不高、智能化程度不足以及鲁棒性差等.为解决这些问题,提出一种改进的YOLOv3卷积神经网络水果分类算法.利用LabelImg工具进行数据标注,把YOLOv3主干网络draknet53替换为DenseNet网络,建立网络层之间的密集连接,增强水果图像的特征信息,实现特征复用,减少计算参数量,强化特征训练,进而训练出一种准确度较高的水果分类模型.经测试,改进的算法对水果分类识别平均准确率达到98%,显著提升了水果分类的准确性.

Fruit classification is of great significance for fruit production,processing,transportation,and self-service sales.Convolutional neural networks efficiently process and classify vast amounts of fruits through multi-level feature learning and automatic feature extraction,showcasing outstanding advantages in fruit classification.However,current fruit classification methods face numerous issues such as excessive reliance on manual labor,suboptimal accuracy,insufficient intelligence,and poor robustness.To address these challenges,an improved YOLOv3 CNN algorithm for fruit classification was proposed.By utilizing the LabelImg tool for data annotation and replacing the YOLOv3 backbone network,draknet53,with the DenseNet network,dense connections between network layers were established.This enhancement reinforced the feature information of fruit images,enabled feature reuse,reduced computational parameters,strengthened feature training,and consequently,resulted in a highly accurate fruit classification model.Tests demonstrated that the improved algorithm achieved an average accuracy rate of 98% in fruit classification recognition,significantly enhancing the precision of fruit sorting.