NEURAL NETWORK EVALUATION OF CHARGE SIMULATION RETINA FOR 3-D IMAGE ACQUISITION AND PROCESSING

Machine vision systems provide the process control information for content, shape, and appearance inspection vital to the customer’s impression of quality product.
Their use is simple, thus, manual labor input is checked.
The image acquisition systems for these machines are limited to 2-D image handling.
Development of machine vision systems that can operate fast enough and achieve 100% inspection and control, yet have the capacity to capture the three dimension detail is anticipated.
Since a charge simulation method algorithm can model input-output relations taking place in 3-D space, a study was conducted using the algorithm to establish the potential of a numerical retina for distinguishing shape and size of objects.

The modeled charge simulation retina (CSR) consisted of sensory and work cells.
Signals at work cells were computed based on images generated by objects of different shapes and sizes and the stimuli applied to sensory cells.
Training of neural networks using the signals resulted in distinct shape indices that compared well with the assigned values.
High correlation coefficients between the computed and assigned size indices existed.
This shows that the charge simulation retina could distinguish the different shapes and sizes.
Thus, it has potential for use as a machine vision system in the agri-industry.

International Symposium on Plant Production in Closed Ecosystems

C. Kanali, H. Murase, Y. Nishiura, H. Takigawa, N. Honami

acquisition, CSR, evaluation, neural networks, processing, 3-D image

https://doi.org/10.17660/ActaHortic.1996.440.69