A biologically inspired system differentiates plants by classifying binaural audio-frequency echo sequences with error detection and resilience.

A biologically inspired system (BIS) classifies binaural audio-frequency echo sequences to recognize two plants that differ mainly in leaf size. The hierarchical BIS system is implemented with two stages. Stage 1 classifies binaural echo spectra with a pair of ear-specific trained single-layer perceptrons. Stage 2 compensates for limited classification accuracy by detecting binaural classification errors and accumulating classification votes by forming a random walk with drift that approaches one of two absorbing boundaries to differentiate the plants. The random walk provides error resilience because a few classification errors are not catastrophic but only increase the sequence length to reach an absorbing boundary. Plant differentiation occurs when the random walk reaches one of two absorbing boundaries within a specified sequence length limit, otherwise plant recognition fails, and the BIS initiates a new plant probe. Two stage 2 approaches are compared. Approach 1 determines binaural echo classification errors before generating a single random walk. Approach 2 generates a random walk for each ear before determining binaural classification consistency. Approach 1 gives more accurate target recognition by generating a random walk having a greater drift to differentiate plants with no errors, 4% failures, and an average of ten views.