LSNet: Identification of copper and stainless steel using laser speckle imaging in dismal surroundings

In unconstrained environments, identification of materials in a non-contact manner is of great necessity. However, most of the current material recognition technologies and their algorithms are contact measurement technologies under restricted conditions. In the current work, we attempt to propose a material recognition solution in the application scenario under harsh conditions. We first set up a specific application scenario to identify the copper and the stainless steel in the dark environment, where the existing material identification technologies are insufficient to carry out material recognition. To accomplish this task, we utilized a laser speckle imaging technique to acquire the speckle images of copper and stainless steel. As we used a laser in the near-infrared band, the whole process of image acquisition was silent. The hardware setup as described above can meet the requirements of multiple special application scenarios such as military reconnaissance. Afterwards, the obtained speckle image was combined with the proposed end-to-end model, i.e., Laser Speckle Network (LSNet). As a result, the classification accuracy of LSNet is 0.963, which is better than other deep learning networks. Also, LSNet is suitable for real-time detection due to the relatively less test time of 4 ms. The experimental results show that the combination of the laser speckle imaging and the proposed LSNet framework can achieve the recognition of the copper and the stainless steel in dismal surroundings. Experimental results show that the combination of the two techniques is feasible. Its unique advantages for dark environments are expected to lead to applications in the military reconnaissance, autonomous driving and other fields.