Hybrid Deep Learning Approach for Wildlife Monitoring from Aerial Forestry Images

This research presents a new combined deep learning system for the effective and reliable identification of plant diseases in complicated agricultural environments. One of the most difficult jobs in forestry is identifying wildlife count and their health conditions along with their daily activities. With the application of machine learning and deep learning techniques, this issue has been resolved. The activities of the wildlife can be monitored automatically, which is advantageous as it reduces the monitoring time and also the fatigue involved in surveillance by the forest rangers. The suggested approach consists of multiple important stages. To begin with, the image quality of the forest lands with the wildlife resource is improved through preprocessing techniques like noise reduction, gamma correction and white balancing. Data augmentation is incorporated to expand the dataset and improve the generalization capacity of the model. Recently, with the advancement of Deep Neural Networks (DNN), the presence and motion tracking comparison has significantly improved the discrimination capability across various applications. Despite the detailed research performed by researchers, the pedestrian tracking performance still needs enhancement due to difficult pedestrian trajectories and frequent obstructions individuals. The proposed Kalman Filter with Convolution Neural Network (KF-CNN) method has been evaluated to identify the location and type of wildlife on multi-class image datasets from the Animals with Attributes (AwA) repository for image analysis. The use of hyperparameter tuning techniques is also implemented to avoid over fitting and improve the overall generalization. This comprehensive approach depicts encouraging results in overcoming challenges faced during the implementation of the Extended Kalman filter (EKF) for wildlife monitoring and detection in forests.