基于宏观图像的木材识别,使用深度学习和转移学习方法
1Department of Computer Engineering, Seydişehir Ahmet Cengiz Faculty of Engineering, Necmettin Erbakan University, Konya, Turkey.
PeerJ
|March 4, 2024
概括
深度学习模型显著改善了森林物种的识别,使识别更快,更容易. ShuffleNet为森林管理和保护工作提供了一种高效,高性能的解决方案.
科学领域:
- 林业科学 林业科学
- 计算机科学 计算机科学
- 人工智能的人工智能是人工智能.
背景情况:
- 准确的森林类型识别对于评估生态,经济和社会效益至关重要.
- 传统的专家观察正在被人工智能 (AI) 等技术进步所增强.
- 深度学习为更快,更有效地识别森林物种提供了潜力.
研究的目的:
- 适应和评估各种深度学习模型用于森林物种识别.
- 使用转移学习评估不同深度网络架构的性能.
- 确定一个轻量级和高效的模型,用于林业的实际应用.
主要方法:
- 使用转移学习调整预先训练的深度网络模型 (RestNet18,GoogLeNet,VGG19,Inceptionv3,MobileNetv2,DenseNet201,InceptionResNetv2,EfficientNet,ShuffleNet),以此进行调整.
- 关于新森林物种数据集的培训和评估.
- 使用指标进行绩效评估:准确性,回忆力,精度,F1得分,特异性和马修斯相关系数.
主要成果:
- 深度网络模型在森林物种识别方面表现出有效性.
- 舒弗利网 (ShuffleNet) 作为一种轻量级的模型,以降低计算需求实现高性能.
- 定制的ShuffleNet实现了与其他模型相比的准确性,突出了其效率.
结论:
- 深度学习模型是促进森林物种识别的强大工具.
- 该研究为森林保护和可持续管理提供了宝贵的见解.
- 像ShuffleNet这样的高效模型可以促进人工智能在林业实践中的更广泛采用.
相关概念视频
01:28Softwoods and Hardwoods
163
Softwoods and hardwoods, derived from different types of trees, are distinguished by their leaf structures and cellular compositions, each serving unique purposes in construction and manufacturing. Softwoods come from cone-bearing trees with needle-like leaves and are predominantly composed of longitudinal cells called tracheids and a smaller proportion of radial cells known as rays. Due to their cellular structure, softwoods are commonly used in construction for structural frames, sheathing,...
163
01:19Introduction to Wood
228
Wood, derived from trees, is a versatile and widely used construction material. Trees feature a trunk surrounded by a protective layer of dead bark. Beneath this outer layer lies the living bark, followed by the cambium, and then the sapwood which transitions into heartwood as it matures. At the center of the trunk is the pith. The age of a tree can be discerned by examining its growth rings, which are concentric bands visible in the trunk's cross-section.
The structural integrity of the...
The structural integrity of the...
228