NSC89-2623-D-110-001 蔣依吾 中山大學資訊工程系 紅外線點目標的檢知法則 Automatic detection of small targets in infrared image sequences containing evolving cloud clutter NSC89-2623-D-110-001 蔣依吾 中山大學資訊工程系
Project description Goal: Automatic, real-time detection of small targets in sequences of thermal infrared images containing evolving cloud clutter PFA (probability of false acceptance) PFR (probability of false rejection) 第一期 (88 第一期 ( 88.7.1 ~ 89.6.30) 尋求不必進行多頻譜紅外線信號The Likelihood Ratio Test及Pre-Whitening處理時不可避免之反矩陣運算及頻域轉換，以計算量較少方式，快速資料融合及做出檢知推論。 第二期 (89.7.1 ~ 89.12.31) 進行多頻譜信號與位元平面間之對應及收集多頻譜紅外線測試信號。
Original image Image with true targets
a single frame of processed output data a single frame of typical input data
Hypotheses: Targets hotter than surrounding -> positive contrast Restricted size and aspect ratio -> temporal and spatial behavior of the background noise, target and clutter on a single pixel basis. Parameters of acquisition: Targets: moving , sensors: either moving or stationary Sensor to scene distance: unknown Speed and altitude of sensor movement: unknown Looking angle of sensor: unknown Ranges of possible target size: wide (2 ~ 200 pixels)
Method
with and
Conventional processing chain Advanced processing chain
Multi-sensor fusion Advantages Availability of complementary information Improved performance to countermeasures Improved performance in adverse conditions -> measurement level fusion: done in raw data feature level fusion: some features extracted from raw data are fused decision level fusion: decisions are drawn from the data of each separate data, then combined to reach a global decision
Measurement level fusion
Original image Bit plane 7
Bit plane 6 Bit plane 5
Bit plane 4 Bit plane 3
Bit plane 2 Bit plane 1
Bit plane 0
Original image Bit plane 7
Bit plane 6 Bit plane 5
Bit plane 4 Bit plane 3
Bit plane 2 Bit plane 1
Bit plane 0