DiversityScanner training and test insect images

The dataset contains the images used in training, validating and testing the CNN used in the DiversityScanner robot. We subsequently used our own images with the detailed camera for the training image data set. We used 5 Malaise trap samples from 3 different locations in Germany near the small towns and villages of Rastatt, Kitzing and Framersbach and 145 3 from the Province of L’Aquila, Italy: Valle di Teve and Foresta Demaniale Chiarano-Sparvera. Thus, a mix of own images from different Malaise trap samples was used. The images for our target taxa were not equally distributed but reflected the abundances of each taxon in the Malaise trap samples. [11]. In total 4.325 color images in 15 classes were used for training, while 1.115 images were used for testing. Photos were taken with a Ximea MQ013CG-E2 camera with a telecentric Lensation TCST-10-40 lens with a magnification of 1x.

本数据集包含用于训练、验证及测试DiversityScanner机器人所用卷积神经网络（Convolutional Neural Network, CNN）的图像。后续我们使用搭载专用相机的自有图像构建训练数据集。我们共采集了来自德国拉施塔特、基青、弗雷马斯巴赫周边乡镇的3处不同地点的5个马氏诱捕器（Malaise trap）样本，以及意大利拉奎拉省泰韦谷、查拉诺-斯帕韦拉公有林区域的145 3个样本。综上，本数据集采用了来自不同马氏诱捕器样本的自有图像混合集。目标分类群的图像分布并不均衡，而是如实反映了马氏诱捕器样本中各分类群的实际丰度[11]。总计15个类别的4325张彩色图像用于模型训练，另有1115张图像用于测试。所有照片均采用Ximea MQ013CG-E2相机搭配Lensation TCST-10-40远心镜头拍摄，放大倍率为1×。