Application of KNN Algorithm Based on Particle Swarm Optimization in Fire Image Segmentation

Abstract

In the field of fire image segmentation, most methods are based on color threshold segmentation, so different thresholds should be set according to different environments. In this process, there are too many manual operations. In order to achieve the automatic segmentation of fire images, a modified KNN segmentation algorithm based on particle swarm optimization is proposed. Firstly, a large number of sample data is cropped, redundant samples are removed, and then an improved KNN is employed to classify image pixels. In this paper, K-Median algorithm is used to cluster samples and reduce the computation of similarity degree in KNN. In this process, Particles Swarm Optimization are adopted to avoid the influence of the initial value of K-Median algorithm on the results. Combined with Euclidean distance and correlation distance, a new similarity function is defined to improve the classification accuracy of KNN algorithm. Experiment results show the proposed algorithm has been improved both in classification accuracy and speed.

Multimedia Image Data Analysis Based on KNN Algorithm

“…On the premise of preserving important information, how to project from high-dimensional data space to low-dimensional space, eliminate or reduce the correlation between data dimensions, reduce data dimensionality, reduce intraclass differences, and increase class spacing. Therefore, it is more convenient, fast, effective, and accurate to extract useful information [ 20 , 21 ]. The CHRIS data preprocessing process is shown in Figure 3 .…”

Dynamic weight HiLo attention network for medical image multiple organ segmentation

“…Despite the high accuracy of Transformer models in tasks such as image classification, object detection and medical image segmentation, as mentioned in the literature, 1 they may suffer from overfitting and reduced generalization performance when dealing with small amounts of training data due to the lack of positional bias in convolutional neural networks. Hence, traditional CNN‐based methods, 3–10 U‐Net‐based models remain a popular choice among researchers for medical image segmentation tasks.…”

“…generalization performance when dealing with small amounts of training data due to the lack of positional bias in convolutional neural networks. Hence, traditional CNN-based methods, [3][4][5][6][7][8][9][10] U-Net-based models remain a popular choice among researchers for medical image segmentation tasks.…”