Temporal Convolution Network and Bi-Directional Long Short-Term Memory -Based Anomaly Detection in Satellite Images for Environmental Monitoring

In satellite anomaly detection, challenges include unbalanced sample distribution, a limited number of fault samples and subtle anomaly characteristics. Data-driven methods for satellite data anomaly detection often face issues such as high false positive rates and limited interpretability, which negatively impact accuracy. This paper, Multi-Head Attention Mechanism, Temporal Convolution Network (TCN), and Bi-Directional Long Short-Term Memory (Bi-LSTM) technique for anomaly detection in satellite data, achieve better accuracy. The MHA provides flexibility in focusing on different feature and parts of the data for satellite data where anomalies might be influenced by complex, long term patterns. The Bi-LSTM network capture context form both past and future time steps, which is enhance the models ability to understand temporal dependencies and detect anomalies in satellites. The TCS are effective in handling temporal data, especially when dealing with varying sequence length, which is essential for detection anomalies in satellite imagery data. The proposed MHA and TCN-LSTM technique is evaluated for detection anomaly, achieving a high accuracy of 0.92, recall of 0.91, F1-score of 0.92 and precision of 0.92 respectively. This performance is compared to existing techniques such as Random Forest (RF) and Convolutional Neural Network (CNN) and LSTM network.