The Significance of Forest Monitoring and the Challenge
The Crucial Role of Forest Patrol
Forest patrol plays an indispensable part in forest monitoring. Currently, the detection of changes in forest patches mainly relies on remote – sensing images and manual visual discrimination. This process demands a vast amount of manpower and time.
The Innovation of an Automated Detection System
A Breakthrough in Technology
Researchers from Central South University of Forestry and Technology, Sanya Academy of Forestry Sciences, Research Institute of Tropical Forestry, Chinese Academy of Forestry, and other institutions have joined forces. They have developed a fully automated forest change – detection system by using unmanned aerial vehicles (UAVs) and artificial intelligence technology. They geometrically correct the images taken by UAVs with the data from the position and attitude system. Then, they utilize convolutional neural networks to extract the forest boundaries. By comparing these with historical data, they can identify areas of forest reduction. After that, the average boundary distance algorithm is employed to eliminate misclassifications, ultimately generating an accurate forest change map.
The Research Area and Initial Testing
The Unique Environment of the Research Site
The research area is located in Tielugang, Hainan Province, in a low – latitude region. This area is deeply influenced by the tropical maritime monsoon climate, with high temperatures throughout the year, mild summers, and warm winters. The average annual temperature is 25.5 degrees Celsius, the average temperature in the coldest month is 20.3 degrees Celsius, and the extreme minimum temperature is 5.1 degrees Celsius. The average annual precipitation is 1255 millimeters. The terrain is a sandbar – lagoon – type harbor, and the soil is sandy.
Testing at Different Flight Heights
The research tested the impact of different flight heights on the corrected images by setting the flight heights to 80 meters and 380 meters. When taking 30 UAV images at a height of about 80 meters for forest change detection, the deep network extracted 72 forest change patches. Among them, 22 patches were correctly identified, while 50 patches were misjudged. By further analyzing the 72 change patches using the average distance method, 61 patches were correctly identified, and 11 patches were misjudged, with an accuracy of 85%. The results indicate that these automated methods can effectively assist in forest change detection and improve monitoring efficiency.
Advanced Testing and the Success of the System
Higher – altitude Testing and Improved Results
On this basis, the researchers took 50 UAV images at a height of 380 meters. They identified 110 forest change patches, among which 32 were correctly identified, and 78 were misjudged. The areas of the misjudged patches were mainly less than 500 square meters, and most of the patches with an area larger than 750 square meters were real – change forest patches. When the average distance threshold was set at 8 meters, the best results were achieved. 104 patches were correctly identified, and 6 patches were misjudged, with an accuracy of 0.95 and a recall rate of 0.94, meeting the requirements of forest detection in China.
The Advantage over Traditional Methods
It is said that traditional forest change detection depends on comparing high – resolution satellite images from different periods. In this study, UAV technology is adopted. Using a single image and RTK positioning data for correction, there is no need for image mosaicking. The system can automatically identify and exclude incorrect forest change areas, achieving an accuracy rate of over 95%. The research paper was recently published in Forests. The research was funded by projects such as the Natural Science Foundation of Hainan Province.
