Improvement and Research of CNN Hard Attention LSTM Image Classification Model Suitable for Water Quality Detection

In order to improve the accuracy and efficiency of water quality detection, firstly, the collected images of water samples are preprocessed and data enhancement is carried out, and the water quality pictures are divided into five classes according to the water color, so as to construct a water quality classification dataset; a water quality detection algorithm combining the Hard Attention mechanism (Hard Attention), Convolutional Neural Networks (CNNs), and Long and Short-Term Memory Networks (LSTMs) is proposed. Because the hard attention mechanism is more interpretable than the soft attention mechanism and is able to focus attention on a specific area in the image, it is better to exclude the complex background that the sample image may contain, such as leaves, stones, water plants, etc. In this experiment, the hard attention mechanism was chosen to handle this task. In addition, because the color difference of water quality images is small, in order to better capture the correlation between adjacent pixels in the image and prevent overfitting problems, the CNN-LSTM model is selected for improvement in this experiment. The paper mainly analyses the way of water quality detection signals according to image signals, and tries to explore the performance of Support Vector Machines (SVM), Convolutional Neural Networks (CNN), Long and Short-Term Memory Networks (LSTM) and their improved CNN-LSTM model after the introduction of Hard Attention Mechanism in the task. The experimental results show that the algorithm in this paper outperforms the other three algorithms in the water quality image classification dataset for the water quality detection image analysis task based on image color.