Multi-scale image classification algorithm based on convolutional neural network (CNN)

In order to solve the problem of insufficient capability of existing models in processing complex multi-scale scenes or data sets containing time dimensions, this paper proposes a new multi-scale feature fusion network (MSFFN), which integrates feature representations from VGGNet, ResNet and InceptionNet. effectively capture multiple layers of information in the input image. In addition, to further enhance the temporal dynamic modeling capability of the model, we integrate the LSTM unit in the MSFFN, so that it can better process the data of video or multi-frame image sequences. The experimental results show that the accuracy, accuracy, recall rate and F1 score of MSFFN are significantly improved compared with the single basic model, whether it is static image or video data containing time dimension. Especially in complex multi-scale scenarios, MSFFN shows excellent performance. By combining the capabilities of spatial feature extraction and temporal dynamic modeling, MSFFN not only improves classification accuracy, but also demonstrates its potential for a wide range of applications, such as object detection, semantic segmentation and other visual tasks. This research provides an effective solution to the problem of multi-scale image classification, and also provides a new idea for processing time series data, which promotes the development of computer vision. Future studies can further optimize the feature fusion mechanism, expand application scenarios, and explore the possibility of cross-modal learning to achieve a more comprehensive scene understanding. We hope that our work will inspire more researchers to pay attention to this field and jointly promote technological progress.