Single-photon avalanche diode(SPAD)image sensors are widely used in direct time-of-flight(D-TOF)imaging,but their ranging performance is often constrained by limited laser power.This article presents a SPAD-based D-TO...Single-photon avalanche diode(SPAD)image sensors are widely used in direct time-of-flight(D-TOF)imaging,but their ranging performance is often constrained by limited laser power.This article presents a SPAD-based D-TOF imaging system that combines a reconfigurable macro-pixel sensor architecture with a lightweight depth completion algorithm to achieve long-range depth imaging with enhanced spatial resolution under low optical power.The proposed sensor adopts a back-side illuminated(BSI)3D-stacked architecture with programmable macro-pixels that enhance detection sensitivity and enable flexible sensitivity–resolution trade-offs.An injection-locked ring-oscillator-based time-to-digital converter(RO-TDC)array achieves a time resolution of 152.5 ps,enabling accurate TOF measurement at an optical power of 10 mW.To compensate for macropixel-induced resolution loss,a probabilistic normalized convolutional neural network(pNCNN)is employed for depth completion using sparse depth inputs only.Experimental results demonstrate that up to 30×effective resolution enhancement of the system can be achieved via the depth completion algorithm without changing the physical resolution of the sensor.Additionally,the proposed system achieves a maximum ranging distance of 90 m and a range-to-power figure-of-merit(FOM)of9 m/mW,which validates the effectiveness of the system.展开更多
冬小麦拔节期叶绿素状况估算对于冬小麦营养诊断非常重要。本文利用无人机遥感平台获取冬小麦拔节期长势遥感信息,提取多光谱植被指数、RGB影像纹理特征及覆盖度信息,基于多元线性回归(Multivariate linear regression,MLR)、随机森林回...冬小麦拔节期叶绿素状况估算对于冬小麦营养诊断非常重要。本文利用无人机遥感平台获取冬小麦拔节期长势遥感信息,提取多光谱植被指数、RGB影像纹理特征及覆盖度信息,基于多元线性回归(Multivariate linear regression,MLR)、随机森林回归(Random forest regression,RFR)构建了冬小麦SPAD值估算模型。分析多光谱植被指数、纹理特征和覆盖度信息,以及相互结合对于冬小麦SPAD值估算的影响。结果表明:多光谱植被指数、纹理特征、覆盖度的结合(2种类型或3种类型参数结合)可以用于冬小麦拔节期SPAD值的估算,而且相较于单类型参数或两类型参数结合,更多类型参数结合提高了冬小麦拔节期SPAD值的估算精度。而基于相同参数利用随机森林构建的冬小麦拔节期SPAD值估算模型精度均高于多元线性回归构建的模型精度。其中,基于3种类型参数构建的冬小麦SPAD值估算模型精度最高,R2为0.78,RMSE为2.08。各类型参数对冬小麦拔节期SPAD值估算精度的影响由大到小依次为多光谱植被指数、纹理特征、覆盖度。其中,多光谱植被指数构建的模型精度与纹理特征构建的模型精度相近(R2和RMSE分别为0.71、2.36及0.70、2.45)。覆盖度虽然对于SPAD值的估算精度提升最小,但结合其他特征可提高冬小麦SPAD值估算精度(对于RFR模型,R2提高0.02~0.03)。多光谱植被指数、纹理特征、覆盖度的结合提高了模型估算精度,为冬小麦拔节期SPAD值快速估算提供了技术参考。展开更多
基金supported in part by the National Key Research and Development Program of China under Grant 2024YFE0201500in part by the National Natural Science Foundation of China under Grant 62334008,Grant 62274154,Grant 62534004,Grant 92464103,Grant 62404218,Grant 62134004。
摘要Single-photon avalanche diode(SPAD)image sensors are widely used in direct time-of-flight(D-TOF)imaging,but their ranging performance is often constrained by limited laser power.This article presents a SPAD-based D-TOF imaging system that combines a reconfigurable macro-pixel sensor architecture with a lightweight depth completion algorithm to achieve long-range depth imaging with enhanced spatial resolution under low optical power.The proposed sensor adopts a back-side illuminated(BSI)3D-stacked architecture with programmable macro-pixels that enhance detection sensitivity and enable flexible sensitivity–resolution trade-offs.An injection-locked ring-oscillator-based time-to-digital converter(RO-TDC)array achieves a time resolution of 152.5 ps,enabling accurate TOF measurement at an optical power of 10 mW.To compensate for macropixel-induced resolution loss,a probabilistic normalized convolutional neural network(pNCNN)is employed for depth completion using sparse depth inputs only.Experimental results demonstrate that up to 30×effective resolution enhancement of the system can be achieved via the depth completion algorithm without changing the physical resolution of the sensor.Additionally,the proposed system achieves a maximum ranging distance of 90 m and a range-to-power figure-of-merit(FOM)of9 m/mW,which validates the effectiveness of the system.
摘要冬小麦拔节期叶绿素状况估算对于冬小麦营养诊断非常重要。本文利用无人机遥感平台获取冬小麦拔节期长势遥感信息,提取多光谱植被指数、RGB影像纹理特征及覆盖度信息,基于多元线性回归(Multivariate linear regression,MLR)、随机森林回归(Random forest regression,RFR)构建了冬小麦SPAD值估算模型。分析多光谱植被指数、纹理特征和覆盖度信息,以及相互结合对于冬小麦SPAD值估算的影响。结果表明:多光谱植被指数、纹理特征、覆盖度的结合(2种类型或3种类型参数结合)可以用于冬小麦拔节期SPAD值的估算,而且相较于单类型参数或两类型参数结合,更多类型参数结合提高了冬小麦拔节期SPAD值的估算精度。而基于相同参数利用随机森林构建的冬小麦拔节期SPAD值估算模型精度均高于多元线性回归构建的模型精度。其中,基于3种类型参数构建的冬小麦SPAD值估算模型精度最高,R2为0.78,RMSE为2.08。各类型参数对冬小麦拔节期SPAD值估算精度的影响由大到小依次为多光谱植被指数、纹理特征、覆盖度。其中,多光谱植被指数构建的模型精度与纹理特征构建的模型精度相近(R2和RMSE分别为0.71、2.36及0.70、2.45)。覆盖度虽然对于SPAD值的估算精度提升最小,但结合其他特征可提高冬小麦SPAD值估算精度(对于RFR模型,R2提高0.02~0.03)。多光谱植被指数、纹理特征、覆盖度的结合提高了模型估算精度,为冬小麦拔节期SPAD值快速估算提供了技术参考。