Effects of 2D and 3D Max Pooling on LSTMFeature Extraction

LSTM is a valuable approach in the context of land cover change detection due to its ability to capture temporal dependencies and long-term patterns in sequential data. Land cover change detection involves analyzing satellite images or time series data to identify and classify changes in land surface features over time. By time step analysis of images, LSTM can capture even small gradual changes in the area. The recurrent nature of LSTM allows it to understand the relationship between neighboring pixels which could also help to fix missing or noise pixel data if any. Lastly, its temporal feature extraction capability helps it to better distinguish between classes. The success of any machine learning model depends upon feature extraction and its ability to generalize the data well. This research focuses on the MaxPooling operation used in the convolution layer of a convolutional neural network. The various advantages of MaxPooling are extensively studied by implementing and comparing the performance of 2D MaxPooling and 3D MaxPooling in an LSTM network. As a result, this can be concluded that MaxPooling not only helps in dimensionality reduction and size invariance issues but also helps in better local feature extraction, and translational invariance, and is robust to noise variations.