Smoke Detection Algorithm Based on Wavelet Transformation and Energy Analysis

摘要

A novel algorithm to detect the forest fire smoke based on the wavelet transformation and high-frequency energy analysis is proposed. The conventional color-based RGB algorithm has the difficulty to distinguish the smoke from other objects with the same color, such as clouds and fogs, so it leads to some false alarms. Facing the shortages of the RGB algorithm, we propose a wavelet-based algorithm to improve the detection method. After been transformed by the wavelet, the original picture will be divided into four sub-images. One of the sub-images can reflect the low-frequency energy information of the original image and the other three can offer the high-frequency energy information of the original information in horizontal, vertical and diagonal directions correspondingly. Using the wavelet coefficients in the four sub-images, the energy value can be calculated. Through this way, the energy feature of the picture can be distracted and the energy parameters in frequency domain can be figured out. By comparing the ratios of high-frequency energy to the whole energy in the pictures of the smoke, cloud, and fog respectively, evident distinctions can be found. Thus, a special range of the smoke's high-frequency energy ratio can be set by large experimental data. By comparing the data obtained from the camera with the range for the smoke, we can make sure whether the object is smoke or something just with the similar color. The experimental results we obtained can accurately distinguish the smoke from the none-smoke images. Thus, this algorithm can improve the traditional color-based RGB algorithm and the false alarm rate can be reduced.