Deep Learning Based Thermal Image Processing Approach for Detection of Buried Objects and Mines

Abstract

Thermal imaging based mine detection technique is widely adopted due it suitability of detecting buried metallic and also non-metallic land mines in battle fields. Accurate mine detection using thermal images depends on thermal contrast between the soil and mine and it is affected by various factors such as the depth of burial; soil properties and attributes, water content in the soil, mine properties; as well as the time of day of image acquisition. With temporal temperature variations of the soil, it is difficult to distinguish and discriminate between the buried object and the background in the thermal image using the conventionally followed binary thresholding approach in gray scale. This paper presents deep learning region convolution based neural network approach to identify the buried objects in thermal images. A region interest selection using a bound box is followed for identifying the buried object in the thermal image.  From the experimental results, it is found that there is temperature variation in the thermal images of the buried objects due to the change in heat carrying capacity of the surround soil. Proposed neural network method showed 90% accuracy in predicting the target locations of buried objects in the thermal images and it can be extended for land mine detection using thermal image processing approach.