Fiber optic-based sensors have recently gained popularity in a variety of industrial areas. Currently, these sensors are employed to measure a variety of physical and chemical characteristics. Among the many configurations investigated, sensors based on Long Period Fiber Gratings (LPFGs) stand out for their simplicity, low-cost, and multiplexing capabilities, which allows for simultaneous monitoring of multiple parameters. These devices are periodic structures engraved in optical fibers that couple propagation modes at certain wavelengths, resulting in resonances that are sensitive to environmental fluctuations. Traditionally, measurements in these sensors were based on power fluctuations or spectrum shifts. However, the possibility for employing higher-order core-excited modes to forecast changes in observed variables has not yet been investigated. This methodology may be practical using signal processing methodologies and deep learning techniques. In this study, we developed and implemented a convolutional neural network (CNN) to predict thermal variations based on the modal distribution in LPFGs. An LPFG with a period of 560 µm and length of 33.6 mm was constructed in a few-mode optical fiber (SM2000) using a CO₂ laser etching technique. To train and verify the CNN-based model, a database of 355 empirically acquired near-field images corresponding to the LP11 propagation modes was used. The images were captured with a WIDY SWIR 640 VS infrared camera and a 980 nm laser. Similarly, the model's hyperparameters were tuned using the computational tool Optuna, which improved its overall performance. The findings show that the constructed deep learning model can predict temperature with 98.5% accuracy over a range of 26°C to 120°C, with a maximum error of 3.77°C. The root mean square error (RMSE) of the forecasts was 0.94°C, indicating that the model was accurate. Finally, the inference time for a batch of 32 images was 0.055 seconds, confirming the effectiveness of the proposed approach.