船只目标检测场景船只检测模型数据

该数据为二次加工数据，主要以船只实摄图片为基础，构建神经网络模型，完成目标检测、分类；通过数据训练得到合适的模型数据。数据可用于船只目标检测等场景。1.以船只实摄图片为基础，构建神经网络模型，完成目标检测、分类。具体做法是搭建含特征提取层、特征交互层、检测分类层的目标检测网络，每一层由若干卷积模块组成，每一个卷积模块由若干模型参数确定；以同样方法构建目标分类网络，与目标检测网络串联，按目标检测-目标分类的顺序搭建完整的目标检测、分类网络；图片被输入到网络后先由目标检测网络给出图片中船只所在的位置信息，再由目标分类网络给出对应位置船只的细分类别信息，完成目标检测、分类。2.通过数据训练得到合适的模型参数，将模型参数二值化后进行切分存储。具体做法是：在模型参数训练过程中，把带有标注的样本图片输入到目标检测、分类网络中，将模型输出结果与正确结果（标注结果）进行比较，将偏差值在网络中进行反向传播，对各层的模型参数进行修正；反复进行此过程，直到模型输出结果达到较高准确度为止；然后对得到的模型参数进行编码，按固定长度进行切分后存储。

This is a secondarily processed dataset. It is mainly based on real-world photographed ship images to construct neural network models for object detection and classification, and obtain appropriate model parameters via data training. This dataset is applicable to scenarios such as ship object detection.
1. Construct neural network models for object detection and classification based on real-world photographed ship images. The specific implementation is as follows: build an object detection network that includes a feature extraction layer, a feature interaction layer, and a detection-classification layer, where each layer consists of multiple convolutional modules, and each convolutional module is defined by several model parameters; then construct an object classification network using the same method, connect it in series with the object detection network, and assemble a complete object detection and classification network in the sequence of "object detection → object classification". After an image is input into the network, the object detection network first outputs the position information of the ships within the image, and then the object classification network provides the fine-grained category information of the ships at the corresponding positions, thereby completing the object detection and classification tasks.
2. Obtain appropriate model parameters via data training, then binarize the model parameters and split them for storage. The detailed steps are: during the model parameter training process, input the annotated sample images into the object detection and classification network, compare the model's output results with the ground truth (annotation results), perform backpropagation of the deviation values in the network to update the model parameters of each layer; repeat this process until the model's output accuracy reaches a sufficiently high level; then encode the obtained model parameters, split them into fixed-length segments and store them.

搜集汇总

数据集介绍

特点

该数据集为船只目标检测场景下的模型数据，包含823条二次加工数据，用于构建神经网络模型进行船只目标检测和分类。数据来源于企业数据，无更新频率，适用于船只目标检测等场景。