In recent years, convolutional neural network (CNN)-based artificial intelligence algorithms have been widely applied to object recognition and image classification tasks. However, the high performance of convolutional neural networks comes at the cost of high-intensity computing and enormous numbers of parameters, which pose substantial challenges to terminal implementations. An end-to-end FPGA-based accelerator is proposed in this work that efficiently processes fine-grained pruned CNNs. A group pruning algorithm with group sparse regularization (GSR) is introduced to solve internal buffer misalignments and load imbalances of the accelerator after fine-grained pruning. A mathematical model of accelerator access and transmission is established to explore the optimal design scale and calculation mode. The accelerator is optimized by designing sparse processing elements and by scheduling the on- and off-chip buffers. The proposed approach reduces the computation of a state-of-the-art large-scale CNN, VGG16, by 86.9% with an accuracy loss on CIFAR-10 of only 0.48%. The accelerator achieves 188.41 GOPS at 100 MHz and consumes 8.15 W when implemented on a Xilinx VC707, making it more energy-efficient than previous approaches.
As a challenging computer vision task, fine-grained visual categorization (FGVC) has received more and more attention. Most fine-grained classification algorithms are data-hungry and time-consuming, on the contrary, humans have the ability to recognize different species with only a small number of samples. For example, a baby can identity different kinds of dogs with only few images. Therefore, in this paper, a fined-grained classification model which is based on meta learning is proposed, and it can be used for more accurate fined-grained classification with only a small number of data. Meta learning uses meta-learners to learn relative tasks so that they can handle with new tasks faster and better when facing new tasks. Specifically, our model uses a task learner to learn various tasks and update the parameters of meta learner at the same time. When facing a new task, meta learner can learn and adapt to new tasks more quickly. The key novelty of our model is an attention module which can enlarge the local area response of image. This model has achieved great results on both CUB-200-2011 and Stanford dogs, and the accuracy of this model has achieved 81.74% with only one image of CUB-200-2011.