Multi-level learning features for automatic classification of field crop pests

The classification of pest species in field crops, such as corn, soybeans, wheat, and canola, is still challenging because of the tiny appearance differences among pest species. In all cases, the appearances of pest species in different poses, scales or rotations make the classification more difficult. Currently, most of the classification methods relied on hand-crafted features, such as the scale-invariant feature transform (SIFT) and the histogram of oriented gradients (HOG). In this work, the features of pest images are learned from a large amount of unlabeled image patches using unsupervised feature learning methods, while the features of the image patches are obtained by the alignment-pooling of low-level features (sparse coding), which are encoded based on a predefined dictionary. To address the misalignment issue of patch-level features, the filters in multiple scales are utilized by being coupled with several pooling granularities. The filtered patch-level features are then embedded into a multi-level classification framework. The experimental results on 40 common pest species in field crops showed that our classification model with the multi-level learning features outperforms the state-of-the-art methods of pest classification. Furthermore, some models of dictionary learning are evaluated in the proposed classification framework of pest species, and the impact of dictionary sizes and patch sizes are also discussed in the work.