Thermal safety of Li-ion Batteries is an important issue in their applications in electric vehicles and grid-tied energy storage. Thermal monitoring of Li-ion Batteries is crucial in preventing thermal failures and prolong lifetime. However, the existing temperature monitoring method are sensitive to corrosion and electromagnetic radiation, without considering the aging factor and non-electric features leading to low accuracy for full life cycle thermal monitoring of batteries under different operating conditions. To tackle the challenges, this paper presents a thermal model using convolutional neural networks incorporating FBG sensors. In order to more accurate, the attribute information fusion of non-electric variables (wavelengths) and the electric variables (voltage, current, state of charge) is developed for battery temperature monitoring in the proposed method. The experimental results based on different cells indicate that the proposed model has desirable accuracy which achieves the root mean squared error around 0.2 \(^\circ \) C on the validation data.