PollenIdentificationAndDetectionBasedOnDeepLearning

1. By combining Felzenszwalb algorithm with neural network, unsupervised semantic segmentation is realized and target location is successfully achieved without label of data;
2. Designed an optimized up-sampling method and the effect is remarkable;
3. The precision of pollen identification reached 97.1% by fine-tuning the model. The results of the independently compiled network which is inspired by InceptionV3 are relatively successful, and the accuracy reaches 95.68%. Under this data set, its convergence speed, computational efficiency and parameter number are basically better than those of other tested networks.