边缘AI的霜雪识别系统设计与实现

为提高霜雪识别系统的识别率,开展了基于边缘AI的霜雪识别系统设计。综合海思嵌入式处理器Hi3559A高算力、低功耗特点,硬件部分采用模块复用的方法降低功耗和资源消耗,软件部分,从模型训练、模型量化、模型部署入手对MobileNetV2图像分类网络和ISP自适应处理算法进行改进,最终识别率99.7%,预处理时间0.3 s,图像分类时间0.8 s,模组平均功率2 W,鲁棒性好、性能稳定。由此可见Hi3559A智能相机模组可有效纠正雪深探测仪的疑误数据。

Palmprint recognition method by combining weighted adaptive MULBP and 2DPCA

In order to improve the recognition rate of the frost and snow recognition system, a frost and snow recognition system powered by edge AI was designed. The design comprised the character of Hi3559 A both excellent computation and low power consumption. The hardware used reduce power and resource consumption by Module reuse. For the software, by integrating the characteristics of the high computing power of Hisi, embedded processors Hi3559 A and IMX334, and starting from model training, model quantization and model deployment, the author improved the MobileNetV2 image classification network and ISP adaptive processing algorithm. The final recognition rate reached 99.7%, pre-processing time reached 0.3 s, image classification time reached 0.8 s, module average power was 2 W, and has its properties being robustness and stable. Hence, the Hi3559 A intelligent phase machine module can effectively correct the inaccuracy of the data from the snow depth detector.