中文题名: | 基于FCOS的野生动物图像目标检测 |
姓名: | |
保密级别: | 公开 |
论文语种: | 中文 |
学科代码: | 070101 |
学科专业: | |
学生类型: | 学士 |
学位: | 理学学士 |
学位年度: | 2022 |
学校: | 北京师范大学 |
校区: | |
学院: | |
第一导师姓名: | |
第一导师单位: | |
提交日期: | 2022-05-23 |
答辩日期: | 2022-05-23 |
外文题名: | Wildlife Detection Based on FCOS Network |
中文关键词: | |
外文关键词: | |
中文摘要: |
野生动物是生态环境重要的组成部分,对野生动物进行实时的监测保护有着十分重要的意义,在红外自动相机广泛普及的今日,获取动物的图像数据已不成问题,但如何在减少人工成本和时间成本的前提下对环境背景复杂的动物监控图像进行批量处理成为了当下的重点问题。为利用深度学习解决上述问题,本文对卷积神经网络的一般结构进行了介绍,并对比分析了各类算法的特点和优劣,最终选择了单阶段无锚框的FCOS网络模型来用于野生动物目标检测。实验结果表明,FCOS网络模型在检测效果好的同时也有着较强的抗干扰能力,面对环境各异大小不一的物种,检测结果的平均精度达到了0.95,这为野生动物目标检测自动化提供了相应理论和模型技术支撑。
﹀
|
外文摘要: |
Wildlife is an important part of ecological environment, and it is very important to monitor and protect wildlife in real time. With the widespread popularity of infrared automatic cameras, it is no longer a problem to obtain animal image data, but how to batch process animal monitoring images with complex environmental backgrounds while reducing labor cost and time cost has become a key issue nowadays. In order to solve the above problem by using deep learning algorithm, this paper introduces the general structure of convolutional neural network, and compares and analyzes the characteristics and advantages of various algorithms. Finally, a single-stage anchor-free FCOS network model is selected for wildlife target detection. The results of the experiment show that the FCOS network model has a good detection effect and also has a strong anti-interference ability, and the average precision of the detection results reaches 0.95 in the face of species with different environments and sizes, which provides the corresponding theoretical and technology support for the automation of wildlife target detection.
﹀
|
参考文献总数: | 12 |
插图总数: | 9 |
插表总数: | 6 |
馆藏号: | 本070101/22139 |
开放日期: | 2023-05-23 |