Objective: The computer image processing and neural network technology are applied to diagnose the thermal energy of boiler plants, i. e., the flame combustion diagnosis, to verify their effectiveness and superiority. Methods: First, the YD-NQ type endoscopic high temperature video acquisition system is used to collect the images of flame combustion. Second, the images are pre-processed by the gray-scale method and the median filtering method. Then the flame combustion parameter features are extracted. The neural network algorithm is improved, and the boiler combustion model based on the improved neural network algorithm is established. Therefore, the combustion decision base is obtained. Finally, the improved neural network model is compared with the traditional neural network model and the 5-4 model to verify its validity. Results: The experiments have found that the improved neural network model is superior to the traditional neural network model. Meanwhile, its accuracy rate and confidence are relatively higher than those of the traditional model. In addition, a single sample also consumes shorter running time, which is 0.0075 seconds. Comparing with the 5-4 model, the improved neural network model has certain advantages, i. e., its accuracy rate and confidence are relatively higher, which are, respectively 91.28% and 96.69%, however, a single sample consumes longer running time than the 5-4 model. Conclusion: The experimental research has found that the application of computer image processing and neural network technology to the thermal energy diagnosis of boiler plants can effectively determine the stability of flame combustion, timely understand the state of flame combustion, and thus diagnose the thermal energy. Therefore, they have values for applications.
AK-DL: A Shallow Neural Network Model for Diagnosing Actinic Keratosis with Better Performance than Deep Neural Networks
