Forest cloud image segmentation method, storage medium and equipment

Abstract

The invention discloses a forest cloud and fog image segmentation method, a storage medium and equipment, and belongs to the field of computer vision and forest fire prevention video monitoring. The method comprises the steps of obtaining a forest cloud image; then, forming a semantic segmentation network model by connecting a lightweight neural network with an encoding module and a decoding module in series to process the forest cloud and fog image, carrying out forward calculation on the forest cloud and fog image in the semantic segmentation network model for image segmentation, and outputting a mask graph of a cloud and fog area in the forest cloud and fog image. The method can overcome the defects of low precision, poor generalization ability and the like of a traditional dark channelmethod, and has higher precision while quickly segmenting the forest cloud and fog image, so that a forest fire prevention monitoring system can be effectively assisted in forest smoke and fire detection. Meanwhile, the invention discloses a storage medium for forest cloud image segmentation and equipment comprising the storage medium, and the storage medium and the equipment can be directly deployed in engineering and are used for forest cloud image segmentation.

Description

technical field [0001] The invention belongs to the field of computer vision and forest fire prevention video monitoring, and relates to a method, storage medium and equipment for forest cloud image segmentation. Background technique [0002] Forest fire is a kind of destructive and difficult to eliminate natural disasters. In the world, tens of thousands of casualties are caused by forest fires every year, and property losses are even more inestimable. For example, in recent years, California forest fires in the United States, Indonesia forest fires, Amazon forest fires have aroused widespread concern around the world. These forest fire events that have attracted worldwide attention have not only caused very serious casualties and property losses, but also had a serious impact on the earth's environment. my country's forest area is vast, and the loss caused by forest fires is very serious every year. Due to weather, climate and other reasons, once a forest fire occurs, it...

AI Technical Summary

Problems solved by technology

However, there is an obvious shortcoming in the video-based forest smoke detection method. Since the forest smoke detection algorithm mainly detects the smoke in the early stage of the fire, and there is a certain similarity in the shape characteristics between the fire smoke and the cloud in the forest, it is easy to cause false alarms. , therefore, in order to improve the reliability of the forest fire detection algorithm, the cloud area is segmented and shielded before the detection, so as to reduce the false alarm rate of the algorithm and improve the generalization ability of the algorithm

[0003] In the prior art, the dark channel method is usually used for forest cloud segmentation, but when cloud segmentation is based on the dark channel method, it relies on artificially set thresholds, and can only segment roughly cloud areas. When the scene changes, the threshold needs to be adjusted to be correct. Segmentation, so the accuracy rate is relatively low, and the generalization ability and robustness are poor

For example, when using the dark channel method to segment a single frame of forest cloud image, the time required is about 20ms, and the segmentation accuracy is about 80%. Segmentation requires high segmentation accuracy at the same time. If the accuracy is low, it will cause false alarms, so that the fire cannot be detected in time. Compared with the time cost, the loss caused will be very huge

[0004] In summary, the current forest cloud segmentation mainly has the disadvantages of low segmentation accuracy, difficulty in balancing accuracy and efficiency, and the need to modify the threshold according to the forest scene, and poor applicability.

Images