Development of durian flesh slicing machine for local agriculturists

Abstract

Exported cut orchidsinflorescencesneed to be inspected to prevent infestation, especially from common cutworms, orchid midges and cotton thrips. Such an inspection task requires specialized skills; prolonged inspection nevertheless results in fatigue and hence lower inspection efficiency. An automatic conveyor-belt inspection machine employing image processing technology was therefore developed in the present research. Orchids inflorescences were transported into the photography chamber via a conveyor belt; photos were taken using several cameras installed at several angles. Image data were analyzed via convolutional neural network (CNN) to classify images into those consisting of common cutworms, orchid midges, cotton thrips and those with no insect pests. Based on the classification results, the image classification efficiencies were 78.6% for common cutworms, 68.0% for orchid midges, 39.8% for cotton thrips and 39.1% for non-pest images. Classification errors were due to the excessively smaller sizes of insects, especially in the cases of orchid midges and cotton thrips. These resulted in images with and without insect pests were similar and not distinguishable by the model. Classification accuracy of the machine was not yet at a level suitable for industrial use. The model must be further developed to allow detection of small insect pests with higher accuracy.