Abstract:
Aerial scenes, unlike nature scenes, have many objects crowded on the surface in bird’s view, requiring more discriminative features and local semantics.
Most convolution neural networks (ConvNets) used for scene classification depict global semantics of images and lose low- and mid-level features, especially as the model gets deeper. This paper proposes a multiple-instance densely-connected ConvNet (MIDC-Net) for aerial scene classification to address these issues.
Aerial scene classification is a multiple-instance learning problem to study local semantics. Our classification model has an instance-level classifier, multiple instance pooling, and a bag-level classification layer.
A simplified dense connection structure preserves features from different levels in the instance-level classifier. Convolution features become instance feature vectors. Trainable attention-based multiple instance pooling follows.
It outputs the bag-level probability and highlights scene label-relevant local semantics. Finally, bag labels supervise this multiple instance learning framework with our bag-level classification layer. Our method outperforms many state-of-the-art methods with fewer parameters on three aerial scene benchmarks.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
