Optimized wild animal detection technique through effective features selection and classification by using convolutional gated recurrent network

The camera trapping framework is a technique for monitoring wild animals in the environment without disturbing. It employs autonomously triggered cameras to acquire massive amounts of data for wildlife research. Conversely, camera trapping often produces a large number of false positives along with poor-quality images, and the opportunities for ongoing animal observation are thus extremely limited. This study proposes a deep learning-based animal behaviour detection approach to address these problems. Pre-processing, feature extraction, feature selection, and detection are the four most important processes in the proposed method. The input images are initially gathered from a publicly accessible dataset. The gathered raw input photos are pre-processed to enhance the image quality by using key frame extraction, data cleaning and frame resizing. After the effective pre-processing, the Enhanced Haar wavelet model is utilized to extract the features. Next, the Hybrid Chimp-Zebra Optimization (HCZO) method is utilized to remove irrelevant traits by selecting optimal features in order to minimize the complexity of computation. An adaptable Chameleon-based convolutional long short-gated Recurrent network (AC_CLSGRN) is utilized to recognize wild animals. The proposed detection model is optimized using an adaptive chameleon optimization (ACO) technique. The proposed approach obtained 98.25% accuracy in wild animal detection with a better performance rate and low energy consumption.