AI機器視覺即時影像分析

AI Seeler Real-time Image Analysis

智能尋標器旨在探討基於深度學習雲層辨識與船隻辨識偵測之結果來進行衛星飛行路徑取像姿態規劃，初期完成將雲層辨識Darknet YOLO v3訓練模型與衛星飛行路徑取像姿態規劃兩者整合實現在嵌入式系統NVIDIA TX2GPU上以評估驗證其可行性。最後將訓練模型轉換成Caffee模型並移植到FPGA平台執行。

The AI seeker aims to study the imaging attitude plans of satellite flight path based on the results of deep learning cloud identification and ship identification detection. Initially, the cloud recognition Darknet YOLO v3 training model and the imaging attitude plans of satellite flight path are integrated. It is implemented on the embedded system NVIDIA TX2GPU to evaluate and verify its feasibility. In planted, the training model is finally converted into a Caffee model and to the FPGA platform for execution. At present, the cloud detection accuracy of the cloud recognition training model is higher than 90%, and the speed is sufficient to meet the needs of satellite imaging. It can program the flight attitude plan before imaging, and the flight attitude plan will also reach the optimum solution in the whole region. The ship recognition training model is incorporated.