An SSD-MobileNet Acceleration Strategy for FPGAs Based on Network Compression and Subgraph Fusion
Over the last decade, various deep neural network models have achieved great success in image recognition and classification tasks. The vast majority of high-performing deep neural network models have a huge number of parameters and often require sacrificing performance and accuracy when they are de...
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-12-01
|
| Series: | Forests |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1999-4907/14/1/53 |
| _version_ | 1797442414718222336 |
|---|---|
| author | Shoutao Tan Zhanfeng Fang Yanyi Liu Zhe Wu Hang Du Renjie Xu Yunfei Liu |
| author_facet | Shoutao Tan Zhanfeng Fang Yanyi Liu Zhe Wu Hang Du Renjie Xu Yunfei Liu |
| author_sort | Shoutao Tan |
| collection | DOAJ |
| description | Over the last decade, various deep neural network models have achieved great success in image recognition and classification tasks. The vast majority of high-performing deep neural network models have a huge number of parameters and often require sacrificing performance and accuracy when they are deployed on mobile devices with limited area and power consumption. To address this problem, we present an SSD-MobileNet-v1 acceleration method based on network compression and subgraph fusion for Field-Programmable Gate Arrays (FPGAs). Firstly, a regularized pruning algorithm based on sensitivity analysis and Filter Pruning via Geometric Median (FPGM) was proposed. Secondly, the Quantize Aware Training (QAT)-based network full quantization algorithm was designed. Finally, a strategy for computing subgraph fusion is proposed for FPGAs to achieve continuous scheduling of Programmable Logic (PL) operators. The experimental results show that using the proposed acceleration strategy can reduce the number of model parameters by a factor of 11 and increase the inference speed on the FPGA platform by a factor of 9–10. The acceleration algorithm is applicable to various mobile edge devices and can be applied to the real-time monitoring of forest fires to improve the intelligence of forest fire detection. |
| first_indexed | 2024-03-09T12:41:30Z |
| format | Article |
| id | doaj.art-0c4cb984069f4f0a81127287718c1748 |
| institution | Directory Open Access Journal |
| issn | 1999-4907 |
| language | English |
| last_indexed | 2024-03-09T12:41:30Z |
| publishDate | 2022-12-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Forests |
| spelling | doaj.art-0c4cb984069f4f0a81127287718c17482023-11-30T22:16:42ZengMDPI AGForests1999-49072022-12-011415310.3390/f14010053An SSD-MobileNet Acceleration Strategy for FPGAs Based on Network Compression and Subgraph FusionShoutao Tan0Zhanfeng Fang1Yanyi Liu2Zhe Wu3Hang Du4Renjie Xu5Yunfei Liu6College of Information Science and Technology, Nanjing Forestry University, Nanjing 210037, ChinaCollege of Information Science and Technology, Nanjing Forestry University, Nanjing 210037, ChinaCollege of Information Science and Technology, Nanjing Forestry University, Nanjing 210037, ChinaCollege of Information Science and Technology, Nanjing Forestry University, Nanjing 210037, ChinaCollege of Information Science and Technology, Nanjing Forestry University, Nanjing 210037, ChinaDepartment of Computing and Software, McMaster University, Hamilton, ON L8S 4L8, CanadaCollege of Information Science and Technology, Nanjing Forestry University, Nanjing 210037, ChinaOver the last decade, various deep neural network models have achieved great success in image recognition and classification tasks. The vast majority of high-performing deep neural network models have a huge number of parameters and often require sacrificing performance and accuracy when they are deployed on mobile devices with limited area and power consumption. To address this problem, we present an SSD-MobileNet-v1 acceleration method based on network compression and subgraph fusion for Field-Programmable Gate Arrays (FPGAs). Firstly, a regularized pruning algorithm based on sensitivity analysis and Filter Pruning via Geometric Median (FPGM) was proposed. Secondly, the Quantize Aware Training (QAT)-based network full quantization algorithm was designed. Finally, a strategy for computing subgraph fusion is proposed for FPGAs to achieve continuous scheduling of Programmable Logic (PL) operators. The experimental results show that using the proposed acceleration strategy can reduce the number of model parameters by a factor of 11 and increase the inference speed on the FPGA platform by a factor of 9–10. The acceleration algorithm is applicable to various mobile edge devices and can be applied to the real-time monitoring of forest fires to improve the intelligence of forest fire detection.https://www.mdpi.com/1999-4907/14/1/53forest fire monitoringlightweight networkpruningquantizationsubgraph fusion |
| spellingShingle | Shoutao Tan Zhanfeng Fang Yanyi Liu Zhe Wu Hang Du Renjie Xu Yunfei Liu An SSD-MobileNet Acceleration Strategy for FPGAs Based on Network Compression and Subgraph Fusion Forests forest fire monitoring lightweight network pruning quantization subgraph fusion |
| title | An SSD-MobileNet Acceleration Strategy for FPGAs Based on Network Compression and Subgraph Fusion |
| title_full | An SSD-MobileNet Acceleration Strategy for FPGAs Based on Network Compression and Subgraph Fusion |
| title_fullStr | An SSD-MobileNet Acceleration Strategy for FPGAs Based on Network Compression and Subgraph Fusion |
| title_full_unstemmed | An SSD-MobileNet Acceleration Strategy for FPGAs Based on Network Compression and Subgraph Fusion |
| title_short | An SSD-MobileNet Acceleration Strategy for FPGAs Based on Network Compression and Subgraph Fusion |
| title_sort | ssd mobilenet acceleration strategy for fpgas based on network compression and subgraph fusion |
| topic | forest fire monitoring lightweight network pruning quantization subgraph fusion |
| url | https://www.mdpi.com/1999-4907/14/1/53 |
| work_keys_str_mv | AT shoutaotan anssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT zhanfengfang anssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT yanyiliu anssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT zhewu anssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT hangdu anssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT renjiexu anssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT yunfeiliu anssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT shoutaotan ssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT zhanfengfang ssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT yanyiliu ssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT zhewu ssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT hangdu ssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT renjiexu ssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion AT yunfeiliu ssdmobilenetaccelerationstrategyforfpgasbasedonnetworkcompressionandsubgraphfusion |