apple fruitlet detection （2021）苹果图像识别

基于深度学习的苹果图像识别-文献阅读

apple fruitlet detection （2021）

YOLO V5s-based deep learning

abstract

目标：develop an accurate apple fruitlet detection method with small model size based on a channel pruned YOLO V5s deep learning algorithm.

key method:

1 模型迁移

using transfer learning, a YOLO V5s detection model was built to detect apple fruitlets.

2 简化模型

To simplify the detection model and ensure the detection efficiency, a channel pruning algorithm was used to prune the YOLO V5s model.

用channel pruning algorithm简化模型，在深度学习模型中，每个通道都对应着一个特征图（feature map），去掉对模型贡献不大的通道，减少参数和计算量，提高推理速度

模型效果

A recall, precision, F1 score, and false detection rate of 87.6%, 95.8%, 91.5% and 4.2%, respectively, were achieved; the average detection time was 8 ms per image; and the model size was only 1.4 MB.

要区分直射光 direct sunlight和背光 backlight, 并记录当天天气光照情况，作为参数

框架：采集图像、标记、训练、优化—简化通道算法进行优化

准确性测试，用不同天气和光照的图片测试准确率

简化后和优化后的模型比较，在保证准确度的基础上提高速度

比较不同模型效果：准确率、大小、速度、参数数量

比较有无数据增强的数据集训练出的YOLO V5模型

tips:

准确率（accuracy）、precision 预测精度 、recall 召回率 、f1-score的关系

来源于知乎

准确率(accuracy)=(TP+TN)/(TP+FN+FP+TN)

精确率(precision)=TP/(TP+FP)

你认为的正样本中，有多少是真的正确的概率

召回率(recall)=TP/(TP+FN)

找出的正样本占真实正样本的比例
F1score

最后：The network model was effectively simplified in this work; and in the future, we will focus on improving the apple fruitlet detection accuracy.

思考

相比于准确度，应用上减少内存消耗，提高速率很重要，模型要提高速率，简化参数，保证准确度

原文：Dandan Wang, Dongjian He. Channel pruned YOLO V5s-based deep learning approach for rapid and accurate apple fruitlet detection before fruit thinning

https://www.sciencedirect.com/science/article/pii/S1537511021001999

DOI：10.1016/j.biosystemseng.2021.08.015