基于深度学习的树种图像自动识别

首先，在图像库建立时，为增加特征选择多样性，选择树木的树皮和树叶图像，保留自然背景；另 外，考虑到同一树种在不同树龄条件下树皮图像存在差异，因此加入不同树龄的树皮图像，并用胸径指标来表示 树龄大小。其次，对每类树种图像随机挑选１００ 张作为测试集，剩余数据集全部作为训练集。 通过反复试验比 较不同 ＣＮＮ 结构设置、卷积层数量、全连接层层数、学习率等对结果的影响

First, when constructing the image dataset, to enhance the diversity of feature selection, tree bark and leaf images were selected while retaining their natural backgrounds. In addition, considering that bark images of the same tree species vary under different tree ages, bark images of various tree ages were incorporated, with the diameter at breast height (DBH) used as the indicator to quantify the tree age. Second, 100 images were randomly selected from each tree species category as the test set, and the remaining dataset was used as the training set. Finally, repeated experiments were conducted to compare the effects of different CNN architectural settings, the number of convolutional layers, the number of fully connected layers, learning rate, and other relevant factors on the experimental results.

搜集汇总

背景与挑战

背景概述

该数据集隶属于'东北天然次生林抚育更新技术研究与示范'项目，旨在通过深度学习技术实现树种图像的自动识别。它包含基于树皮和树叶图像构建的数据，并采用卷积神经网络（CNN）进行模型训练与测试，以优化识别效果。