Fruit Weight Measurement and Categorization Using Convolution Neural Network

In India, annual fruit production is within the range of 98 million metric tons. Before being exported, all fruits are subjected to look at for internal control purpose and are graded in step with their maturity, size, and presence of a defect. The scale of fruit is usually defined by its mass because it is relatively simple to live. Within the paper, we propose a fruit weight measurement approach named as FWNet using a convolution neural network. The fruit image (input) is processed through the proposed FWNet to anticipate the fruit weight. We use pre-trained weight parameters of the prevailing VGG-16 to initialize the parameters of the proposed FWNet. The proposed FWNet is validated for fruit weight measurement using the COFILAB fruit image dataset. Image processing techniques proposed for the measurement of area and perimeter to map the scale of pomegranates shows the coefficient of determination R2 up to 0.7529. CNN is proposed for evaluating Mean Absolute Error and Mean Square Error; the experimental analysis witnessed the prevalence of the proposed FWNet over the opposite existing approaches like Inception V3, ResetNet and Mobile Net for fruit weight measurement. Comparative analysis is formed between actual and estimated weight, which shows an average accuracy of 95.67 percentages. Thus image processing and CNN combination can provide a beneficiary result for weight estimation and classification of pomegranate consistent with weight.