.png){kind=icon}
.png){kind=icon}
Return
Multidimensional Excretion Perception Using a Laser Dot Matrix Sensor
.png){kind=icon}
PRE
AI
J
Z
L
Y
L
Y
Z
郭
DOI：10.1109/jsen.2026.3705798.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Conventional excretion detection methods often suffer from limited information dimensionality, dependence on multiple sensors, and high system complexity. To address these limitations, this article proposes a multidimensional excretory state perception framework based on a single laser dot matrix sensor (LDMS). The proposed approach actively acquires distance and reflection intensity distributions of the excretion area to achieve simultaneous recognition of excretion presence, type, shape, and volume. A physical simulation dataset covering diverse excretion categories and volume levels is constructed, with data encoded as pseudo-RGB inputs. Subsequently, a lightweight dual-stream multitask network based on MobileNetV3-small (MNV3S) is designed, incorporating transfer learning and a Fusion-MLP module to fuse dual-modal information for the joint classification (JC) of excretion type/shape and volume classification (VC). Experimental results demonstrate that the model achieves accuracies of 100% and 94.43% for coarse-grained JC and VC, and 85.14% and 90.89% for fine-grained tasks, respectively. Furthermore, deployment on the Jetson Xavier NX platform confirms real-time inference capabilities with a latency of 279 ms, validating the effectiveness and practical feasibility of the proposed method.
Keywords:
Excretion care
laser dot matrix sensor (LDMS)
lightweight neural network
multidimensional excretion sensing
multitask learning
Journal
IF:
4.5
Papers:
2.1W
Citations:
7.3W
Organization
Cited Papers