I have data from a row of sensors that sense a body passing over it. The data is of the shape (NS, NT) where NS is the total number of sensors and NT is the total number of time steps. The sensor data at a given time step is not isolated, and the value sensed at one sensor is related to the values sensed at its neighboring sensors. Also, an individual sensor's data is related to its past and present values sensed.
Now, I am using this data to predict some properties of the body passing over the sensors. I use an LSTM model and pass n_sequence data from my raw dataset for training and the corresponding property value of the body. Thus the input feature to the LSTM model is of shape (NS, n_sequence).
My query is the following. Is the LSTM model able to understand the spatial relation in the sensor data along with the temporal relation? If the answer to the above question is negative, then should I use CNN to capture the spatial data first and then use it's output in LSTM. However If I use a 1D CNN, what should be the shape of my output vector?
