Research On Chronic Trauma Classification Based On Deep Ensemble Learning

Acute and chronic wounds are a challenge to healthcare systems around the world and affect the lives of many people each year.Wound classification is a key step in the diagnosis of acute and chronic wounds and aids clinical diagnosis to determine an optimal treatment procedure.The use of computer-aided recognition and analysis of chronic wound images is an emerging direction in the field of artificial intelligence.At this stage,most algorithms for classifying chronic wounds are aimed at distinguishing a single wound from normal skin or at classifying a single wound,which cannot be applied to the actual diagnosis of patients with chronic wounds,and the few models for classifying multiple types of wounds do not work well,making it important to train a model that can classify multiple chronic wounds.Having a high-performance classifier can therefore help experts in the field to classify wounds with less financial and time costs.In this paper,we propose a model called SARNet based on a network structure with a multi-branch topology,and the main contributions can be summarised as follows:1)This paper investigates chronic wound images based on deep integrated learning,and analyses the shortcomings of the improved Grow Net with partial integrated learning algorithms applied on chronic wound multi-classification.2)Based on the experiments and analysis in this paper,a self-attentive embedded residual network model named SARNet is proposed in this paper.The model is based on a multi-branch topological network structure,allowing each branch to apply different convolutional kernels to obtain different perceptual fields and learn deeper features,and the SARNet model achieves an accuracy of 80.87% for six classifications in a dataset of 1777 chronic wounds.3)The SARNet model demonstrated in this paper achieved excellent performance in multi-classification based on chronic wounds,with a further 0.55% improvement in the ESARNet effect after the addition of integrated learning.Experiments showed that the ESARNet model achieved 81.42% accuracy for six classifications in a dataset of 1777 chronic wounds.The model presents a new idea and is of great value in the field of chronic wound multi-classification,making it possible to classify chronic wounds in practice.