Automatic quality and moisture evaluations using Evolution Constructed Features

• We propose to use a new object classification algorithm called ECO-Features for quality evaluation.
• We evaluate three different ways for efficient training.
• We examine nut quality and date maturity using this improved algorithm.
• Performance is improved differently for different crops.

Many quality evaluation tasks that are complicated and unique to specialty crops are often carried out manually by human experts by visually inspecting product appearances. This labor-intensive process usually depends greatly on experienced workers and lacks verification efficiency. Automating these tasks not only reduces the processing time, improves the verification accuracy, but also reduces the labor costs. This study is conducted to explore the feasibility of using our recently developed Evolution-Constructed Feature (ECO-Feature) to automatically evaluate the quality of agricultural products. Three feature descriptors are added to the original ECO-Feature algorithm to reduce its dimensionality and improve training efficiency. This paper uses date fruit, cashews, pistachios, and almonds as examples to demonstrate the performance of the proposed algorithm for quality and moisture evaluations. The proposed method is proven accurate, effective and has been implemented and deployed for commercial production for the date industry in the United States.