Ecological Big Data Adaptive Compression Method Combining 1D Convolutional Neural Network and Switching Idea

To address large data transmission requirements and high transmission power consumptions characterizing micro-environment monitoring systems that are commonly used in forest health and safety applications, we propose an ecological big data adaptive switching compression method based on a 1D convolutional neural network (1D CNN). First, to ensure that data samples apply to different compression dictionaries, a 1D CNN is used to classify the samples into two sets according to the characteristics of samples. Subsequently, based on the classification results, the switching factor $S$ is defined, such that the discrete cosine transform (DCT) predefined dictionary and the learning dictionary (K-SVD) can be used to adaptively achieve sparse expression and data compression. Finally, the orthogonal matching pursuit (OMP) algorithm reconstructs the sparse signal. To evaluate the feasibility and robustness of the proposed method, we conduct experiments on four types of data: air temperature (AT), air humidity (AH), soil temperature (ST), and soil humidity (SH). The results indicate that the proposed method, compared to K-SVD and DCT dictionary, exhibits excellent performance for all data samples having smaller sparse error (SE), smaller reconstruction error (RE), and larger compression ratio (CR) at different sparsity levels. In particular, when sparsity $K$ is 16, the reconstructed signal is the most similar to the original signal. In addition, the proposed method reduces power consumption by 79.90%, compared with uncompressed data transmission. Considering four factors, the adaptive switching compression method based on 1D CNN has higher reconstruction accuracy and lower power consumption than using only K-SVD or DCT dictionary.