基于多特征融合和CNN模型的树种图像识别研究

将 6 类常见树种（樟子松、山杨、白桦、落叶松、雪松和白皮松）图像作为研究对象。首先，通过裁剪、 水平翻转、旋转等操作，对原始树种图像集进行数量扩增，并划分为训练集和测试集，建立本次树种识别实验的图像库； 其次，将本文模型设计为 3 路并列网络，分别选取 RGB 图像、HSV 图像、LBP-HOG 图像，从图像像素、色彩、纹理和形状 的角度出发，对上述树种图像进行识别。

This research takes images of six common tree species (Pinus sylvestris var. mongolica, Populus davidiana, Betula platyphylla, Larix gmelinii, Cedrus deodara, and Pinus bungeana) as its research subjects. First, the original tree species image dataset was augmented via operations including cropping, horizontal flipping, and rotation, then split into training and test sets to establish the image database for this tree species recognition experiment. Second, the model proposed in this paper is designed as a three-branch parallel network, which selects RGB images, HSV images, and LBP-HOG images respectively, and identifies the aforementioned tree species images from the perspectives of image pixel information, color, texture, and shape.

搜集汇总

背景与挑战

背景概述

该数据集专注于6类常见树种（樟子松、山杨、白桦、落叶松、雪松和白皮松）的图像识别研究，采用多特征融合和CNN模型进行识别。数据集属于东北天然次生林抚育更新技术研究与示范项目，数据格式为文档，适用于植物学领域的深度学习研究。