Mitsuhiko Katahira's research while affiliated with Yamagata University and other places

Tiller number, an important growth parameter for rice cultivation, is still being assessed manually. This work investigated the influence of dataset composition on performance of deep learning models for tiller number estimation in rice. Four datasets were constructed for early tillering, active tillering, and maximum tillering by applying the concepts of mixed varieties, class balance, and data augmentation. YOLOv4 models were trained to estimate tiller numbers using each constructed dataset. Then their performance was evaluated. Results demonstrated that the models trained with datasets created using a combination of mixed variety, class balance, and augmentation showed the best performance for estimating the tiller number at the three tillering stages with a mAP range of 68.8–86.4 %.

Cattle manure is scored to indicate cattle health. Scoring is done by visual observation related to images and appearance features. Nevertheless, the score standard remains obscure. Few reports have described numerical indexes for this score. This study examined quantification of manure characteristics, which is the basis of cattle health judgment to reduce burdens on livestock managers and to facilitate skill acquisition. We investigated cattle manure moisture characteristics by measuring the moisture contents of manure sampled from Japanese black calves. Subsequently, we verified the classification accuracy of manure images based on manure moisture characteristic using deep learning object detection. The range of manure moisture contents was 75.7–93.8%. Manure with moisture contents of 89–91% is spread widely on the bedding. Manure with moisture contents of 92–94% is in a liquid state. Therefore, this manure went under the bedding （rice husk）. The bedding covered the manure. Scores were divided into three levels with moisture content of 6%, four levels with 5%, five levels with 4%, and six levels with 3%. Then, the AI models were made. The F1 score of the AI model was 0.80 for three classification levels, 0.73 for four levels, 0.62 for five levels, and 0.53 for six levels. The F1 score of the AI model for three classification level and four level were significantly higher than the AI model for five level and fix level. So, when the manure score classifies by deep learning, three classification level, and four level are effective.

The monitoring of rice tiller number is one of the most tedious and the time consuming task of rice cultivation. In this work, we use deep learning as an alternative method to estimate tiller number from images captured by a field robot at three tillering stages: early stage, active stage, and maximum stage, for the two Japanese rice varieties of Fukuhibiki and Haenuki. Three types of YOLOv4 models were trained to estimate the tiller number: models aimed at estimating actual tiller numbers, models trained on classes of grouped tiller numbers, and models trained with classes based on a tiller number histogram. In the experiments, the tiller number histogram based models achieved the highest scores of mAP at the three tillering stages of early, active, and maximum: 62.3, 67.5, and 73.5 for Fukuhibiki variety, 61.3, 63.5, and 49.8 for Haenuki variety. ［Keywords］crop sensing, deep learning, field robot, precision agriculture, rice tiller, YOLOv4

This report examined the working properties and machine introduction conditions of the paddy iron-coated direct seeding cultivation system under conditions where the field scale and working systems were different. Two field scales; 1.5 ha and 0.3 ha, were compared by field size, and three working systems; a） included plowing, harrowing and leveling, b） excluding plowing, harrowing and leveling in direct seeding cultivation and c） the transplanting working system excluding plowing, harrowing and leveling. The operating costs of agricultural machines were calculated by measuring the effective field capacity and fuel consumption for each test plot. The effective field capacity of the whole process from soil preparation to harvesting was 2.3 ha h–1 higher for the 1.5-ha field than for the 0.3-ha field, with the calorific value being 2241 MJ ha–1 lower. With respect to working systems, the effective field capacity （calorific value） for the working system that included plowing, harrowing and leveling in 0.3-ha field was 0.8 ha h–1 （3934 MJ ha–1） higher than the working system that did not include these operations and 0.9 ha h–1 （3722 MJ ha–1） higher than the transplanting working system. The operating cost in the paddy direct-seeding work system in the 1.5-ha field was equivalent to the standard operation cost for conventional working systems in Shonai area. The operational cost, based on the findings if general-purpose tractors, plows, harrow, and rotary in 0.3 ha upland fields, for direct-seeding work systems and transplant work systems which exclude plowing, leveling, and leveling, can be below that of operation cost. Therefore, if plowing, harrowing, and leveling are included in the 0.3-ha field, these works should be conducted in the autumn season to increase feasible coverage.

This paper presents discussion of the influences of soil physical properties and power farming system on field vegetable cultivation by different drainage systems on upland fields converted from paddy fields near the Sea of Japan. Conversion of paddy fields with open ditches and mole drains to upland fields （B block） quickly and markedly improved soil physical properties, specifically the water permeability coefficient and uplandization index, compared to another field with an open ditch for soil structure development （A block）. Green soybean yields of first-year cultivation were 935 kg/10 a for A block and 732 kg/10 a for B block. Respective cabbage yields of second cultivation for the first year were 3,178 kg/10 a and 3,684 kg/10 a. Taro yields of second-year cultivation were 3,026 kg/10 a and 2,613 kg/10 a for quickly cultivated type and 2,539 kg/10 a and 2,364 kg/10 a for normal cultivation type. Welsh onion yields of third year cultivation were 4,464 kg/ 10 a and 7,107 kg/10 a. Upland fields converted from paddy near the Sea of Japan should have an open ditch and mole drain before vegetable cultivation. Then, first-year cultivation on upland fields can be expected to introduce green soybean using rotary tilling and a ridge-making implement by up-cut rotary processing, which can increase the pulverization rate and decrease water damage. Next, cultivation should introduce cabbage using rotary tilling and a small ridge-making implement, which might have a high work rate. From the second year, cultivation of upland fields should include taro and welsh onions using the ridge and mulching transplanter cultivation method, and a side-dressing fertilizer applicator ditcher cultivation method for wider use of the plow layer by the improvement of soil physical properties.