Hatchery marked otolith images and classification models

these data contain 250 images of hatchery marked and unmarked otolith images. the images include otoliths with one of four distinct marks. there are 50 images that contain each of the four marks, and 50 images of unmarked otoliths. the images were used to train and test neural networks for use in identifying the marks. the networks were trained on the first thirty images in each class. the remaining twenty images in each class can be used for testing. two trained networks are included: "binarynet" distinguishes marked and unmarked images, and "classnet" classifies marked images. two versions of "binarynet" are available: one trained on the same database as "classnet" and a second iteration stored in a "retrained" directory that was finetuned using adversarial samples selected from the training images by "classnet." finally, a set of software utilities written in python are included that show how the networks were trained and process the images for classification by the networks.

本数据集包含250张孵化标记与未标记的耳石图像。这些图像中耳石上具有四种独特的标记之一。其中包含每个标记的50张图像，以及50张未标记的耳石图像。这些图像被用于训练和测试神经网络，以便用于识别标记。每个类别的神经网络均在每个类别的前三十张图像上进行训练。每个类别剩余的二十张图像可用于测试。包含两个训练好的网络：“binarynet”区分标记与未标记的图像，“classnet”对标记图像进行分类。两种版本的“binarynet”可供使用：其一与“classnet”在相同的数据集上训练，其二存储在“retrained”目录中，通过使用“classnet”从训练图像中选取的对抗样本进行了微调。最后，还包括了一套用Python编写的软件工具，展示了网络的训练过程以及如何对图像进行分类处理。