The advances in developing more accurate and fast smoke detection algorithms increase the need for computation in smoke detection, which demands the involvement of personal computers or workstations. Better detection results require a more complex network structure of the smoke detection algorithms and higher hardware configuration, which disqualify them as lightweight portable smoke detection for high detection efficiency. To solve this challenge, this paper designs a lightweight portable remote smoke front-end perception platform based on the Raspberry Pi under Linux operating system. The platform has four modules including a source video input module, a target detection module, a display module, and an alarm module. The training images from the public data sets will be used to train a cascade classifier characterized by Local Binary Pattern (LBP) using the Adaboost algorithm in OpenCV. Then the classifier will be used to detect the smoke target in the following video stream and the detected results will be dynamically displayed in the display module in real-time. If smoke is detected, warning messages will be sent to users by the alarm module in the platform for real-time monitoring and warning on the scene. Case studies showed that the developed system platform has strong robustness under the test datasets with high detection accuracy. As the designed platform is portable without the involvement of a personal computer and can efficiently detect smoke in real-time, it provides a potential affordable lightweight smoke detection option for forest fire monitoring in practice.

To date, most existing forest fire smoke detection methods rely on coarse-grained identification, which only distinguishes between smoke and non-smoke. Thus, non-fire smoke and fire smoke are treated the same in these methods, resulting in false alarms within the smoke classes. The fine-grained identification of smoke which can identify differences between non-fire and fire smoke is of great significance for accurate forest fire monitoring; however, it requires a large database. In this paper, for the first time, we combine fine-grained smoke recognition with the few-shot technique using metric learning to identify fire smoke with the limited available database. The experimental comparison and analysis show that the new method developed has good performance in the structure of the feature extraction network and the training method, with an accuracy of 93.75% for fire smoke identification.