Abstract
In the context of visual perception, the optical signal from a scene istransferred into the electronic domain by detectors in the form of image data,which are then processed for the extraction of visual information. In noisy andweak-signal environments such as thermal imaging for night vision applications,however, the performance of neural computing tasks faces a significantbottleneck due to the inherent degradation of data quality upon noisydetection. Here, we propose a concept of optical signal processing beforedetection to address this issue. We demonstrate that spatially redistributingoptical signals through a properly designed linear transformer can enhance thedetection noise resilience of visual perception tasks, as benchmarked with theMNIST classification. Our idea is supported by a quantitative analysisdetailing the relationship between signal concentration and noise robustness,as well as its practical implementation in an incoherent imaging system. Thiscompute-first detection scheme can pave the way for advancing infrared machinevision technologies widely used for industrial and defense applications.