Target detection, object classification and semantic segmentation of UAV remote sensing images based on PyTorch deep learning

With the gradual upgrade of UAV automation capabilities,It is widely used in various fields,如航拍、农业、植保、灾难评估、救援、测绘、电力巡检等.But at the same time due to the low flying altitude of the drone、Get more target types、Factors such as complex environment and other factors make the processing of data obtained by drones more and more complicated.最近借助深度学习方法,Impressive results have been achieved for drone target recognition based on convolutional neural networks.深度卷积网络采用“端对端”的特征学习,Feature extraction through multiple layers,It reveals nonlinear features hidden in the data,能够从大量训练集中自动学习全局特征,This is also an important reason for its success in automatic target recognition in UAV images,也标志特征模型从手工特征向学习特征转变.同时,以PyTorchDeep learning platforms such as these also provide procedural frameworks for using convolutional neural networks.However, the mathematical models and computer algorithms involved in convolutional neural networks are very complex、运行及处理难度大,It is not easy to master various deep learning platforms.

【专家】：Dr. Chen,具有资深的技术底蕴和专业背景,And have been engaged in geography for a long time/遥感大数据,机器/深度学习,人工/Research work on brain-like intelligence,Many achievements have been published in top academic journals in the field.

/// 基 于 PyTorch 深 度 学 习 无 人 机 遥 感 影 像 目 标 检 测、地 物 分 类 及 语 义 分 割 实 践 技 术 应 用

Detailed explanation of deep convolutional network knowledge
1.Paradigms and problems of deep learning in drone image recognition
2.深度学习的历史发展历程
3.机器学习,Basic processing flow for tasks such as deep learning
4.卷积神经网络的基本原理
5.Principles and understanding of convolution operations
6.池化操作,全连接层,以及分类器的作用
7.BPAn understanding of the backpropagation algorithm
8.CNN模型代码详解
9.特征图,卷积核可视化分析

PyTorch应用与实践
1.PyTorch简介
2.动态计算图,静态计算图等机制
3.PyTorch的使用教程
4.PyTorch的学习案例
5.PyTorch的基本使用与API
6.PyTorchAn explanation of the image classification task
案例：
(1)不同超参数,如初始化,学习率对结果的影响
(2)使用PyTorchBuild a neural network and implement the classification of handwritten digits
(3)使用PyTorchModify the model and improve the classification model performance

Convolutional Neural Network Practice and UAV Image Target Detection
1.Basic knowledge of UAV image target detection under deep learning
2.目标检测数据集的图像和标签表示方式
3.Explain the evaluation scheme of the target detection model,包括正确率,精确率,召回率,mAP等
4.讲解two-stage（二阶）检测模型框架,RCNN, Fast RCNN, Faster RCNN等框架的演变和差异
5.讲解 one-stage（一阶）检测模型框架,SDD ,Yolo等系列模型
6.A summary of the development of existing detection models

A case of UAV image target detection task
案例1：
（1）一份完整的Faster-RCNN The target detection of UAV images is realized under the model
（2）讲解数据集的制作过程,包括数据的存储和处理,并使用PyTorch加载数据集
（3）数据集标签的制作
（4）模型的搭建,组合和训练
（5）检测任数据集在验证过程中的注意事项
案例2：Plant identification and statistics from drone images

Deep Learning and UAV Image Segmentation Tasks
1.Basic concepts of UAV image segmentation task under deep learning
2.讲解FCN,SegNet,U-net等模型的差异
3.A summary of the development of segmentation models
4.Differences between drone image segmentation tasks and image segmentation
5.Considerations in the task of UAV image segmentation
案例
（1）A case study of drone land cover classification
（2）讲解数据集的准备和处理
（3）A strategy for dividing UAV imagery into small images
（4）模型的构建和训练方法
（5）验证集的使用过程中的注意事项

Semantic classification tasks and deep learning optimization techniques for point cloud data
1.深度学习下的ASL（机载激光扫描仪）Basic knowledge of the task of semantic classification of point cloud data
2.PointNet与PointNet++Basic explanation of the model
案例：
(1)Preprocessing and partitioning of point cloud data
(2)Semantic segmentation of point cloud data
(3)Prediction result analysis of point cloud data
Summary of deep learning related skills：
1.现有几个优秀模型结构的演变原理,包括AlexNet,VGG,googleNet,ResNet,DenseNet等模型
2.从模型演变中讲解实际训练模型的技巧
3.Explain optimization strategies for data
4.Explain the optimization strategy for the model
5.Explain the optimization strategy for the training process
6.Explain optimization strategies for detection tasks
7.Explain optimization strategies for segmentation tasks
8.提供一些常用的检测,分割数据集的标注工具

Additional learning
根据科研或生产实际,Brainstorm deep learning implementations
提供若干附加材料,包括数据集,标签工具、code and learning materials
实例回顾、训练、巩固
答疑与讨论

更多
●【教程】基于PyTorch深度学习遥感影像地物分类与目标检测、分割及遥感影像问题深度学习优化
●【教程】基于PythonDeep learning remote sensing image classification and target recognition、分割
●【教程】PROSAILModel forward simulation and remote sensing extraction code implementation of vegetation parameters
●【教程】高光谱遥感数值建模技术及在植被、水体、Soil information extraction
●【教程】Remote sensing inversion and data assimilation of vegetation parameters
●【教程】“卫星-无人机-地面”The rapid use of remote sensing data and the realization method of ground object content calculation
●【教程】陆面生态水文模拟与多源遥感数据同化的实践技术应用
●【教程】长时间序列遥感数据处理及在全球变化、物候提取、植被变绿与固碳分析、生物量估算与趋势分析等
●【教程】近地面无人机植被定量遥感与生理参数反演实践技术应用
●【教程】无人机遥感在农林信息提取中的实现方法与GIS融合应用
●【教程】InVEST模型在生态系统服务供需、碳中和等领域中的应用及论文写作技能
●【教程】MAXENT模型生物多样性生境模拟与保护优先区甄选、保护区布局优化评估及论文写作技巧